Detailed Project Report for Learning Management System

1. OverviewPDF p.4

This Detailed Project Report defines the business, functional, technical, artificial intelligence, and implementation requirements for the PM-SETU Learning Management System (LMS) as the learner-facing and instructional digital layer of the PM-SETU ecosystem. The system is envisaged as the operational platform through which every participating Industrial Training Institute (ITI) delivers an approved course to an identified learner, records what that learner has done, measures attainment, and presents the resulting progress and credential. The platform provides a unified digital environment for course and content governance, blended learning delivery, assessment, adaptive and AI-assisted learner support, progress tracking, skills and career intelligence, certification, and lifecycle hand-off to opportunity services.

The primary purpose of the Learning Management System is to establish a single, reconcilable learner record across the entire learning lifecycle, beginning from enrollment and course activation through content delivery, practical activity, assessment, progression, completion, credentialing, and transition to on-the-job training, apprenticeship, or employment. Given the program's multi-stakeholder delivery structure involving Central and State authorities, State Project/Skill Vehicle (SPV) entities, Anchor Industry Partners (AIPs), ITIs, trainers, awarding bodies, content partners, and industry organizations that already operate their own learning platforms, the system is intended to create one authoritative learner record without requiring every participating organization to adopt an identical underlying system. The platform strengthens instructional governance by enforcing the approved course version, validating every learning event against the learner and batch to which it belongs, preserving the source of each record, and making outstanding requirements explicit rather than inferred.

This version of the Detailed Project Report extends the LMS with a governed set of artificial intelligence (AI) capabilities validated against current global and Indian market practice. Every AI capability in this report is bound to the same governing principle that already anchors the platform's approach to adaptive learning: AI may inform, recommend, draft, and predict, but it does not silently become the authoritative source for a curriculum requirement, assessment result, a completion state, or a credential. This principle is treated as non-negotiable and is applied consistently across every functional pillar, the architecture, and the non- functional requirements set out below.

1.1 ScopePDF p.4

The PM-SETU Learning Management System provides a centralized digital platform to manage the complete learning lifecycle of the scheme across cluster, institution, trainer, and learner levels. The solution supports course and qualification governance, digital and blended content delivery, theoretical and practical assessment, adaptive and AI-assisted learner support, progress and competency tracking, skills and career intelligence, certification and digital credentials, identity and credential integration, and the learner-facing and institution-facing operational dashboards required to run delivery. It enables the digital management of course versions and batches, including national core content, approved industry modules, practical and workshop activity, and on-the-job training components, together with structured workflows for content review, publication, assessment administration, intervention, and completion.

The scope explicitly includes the interoperability required to make the learner journey work across a heterogeneous delivery landscape. The platform admits a vendor or industry partner that already operates a complete learning management system, consuming its records through a governed interface without requiring the same learner activity to be entered twice; it equally admits a partner that contributes learning content alone, without that partner having to operate any learner-lifecycle system. The scope further includes integration with the approved learner identity and enrollment channel, with the designated examination and awarding services for authoritative results and certificates, with credential services such as DigiLocker and the Academic Bank of Credits where applicable, with the National Career Service (NCS) and Skill India Digital Hub (SIDH) for career and opportunity intelligence, and with apprenticeship and employment services for lifecycle hand-off. The platform exposes validated operational events to the Analytics Portal, but national,

State, and cross-system dashboards are delivered through that portal and are outside this Detailed Project Report.

The scope also includes a governed set of artificial intelligence capabilities, organized under a dedicated AI Governance framework, that assists the learner, the trainer, the assessor, and the institution administrator without displacing the authority already assigned in this report to curriculum approval, formal examination, credential issuance, or institutional decision-making. Where an AI capability is not yet sufficiently mature, evidenced, or governable for a Government of India program at the time of this report, it is identified as a future-phase capability rather than specified for initial delivery, and two capabilities identified during market research - emotion or affect recognition, and fully autonomous grading or credentialing agents - are explicitly excluded from scope on governance and legal grounds set out in Section 6.8.

1.2 Key AssumptionsPDF p.5

• ●Learner identity will be resolved against an approved PM-SETU enrollment and identity record before

any learning activity is created, and staff and partner users will be provisioned against approved roles, institutions, and organizations prior to deployment.

• ●Formal examination results, certificate identifiers, and statutory qualification records will remain

authoritative in the designated DGT, NCVT, or other awarding services. The Learning Management System references and presents these records; it does not generate a competing authoritative record, and no AI capability specified in this report alters this position.

• ●External systems, including the identity channel, awarding and examination services, credential

services, vendor learning platforms, the National Career Service, Skill India Digital Hub, and opportunity services, will provide the required interfaces, data-exchange mechanisms, and service availability to enable governed integration.

• ●Where an industry partner operates a connected learning platform, each learning event will retain

one authoritative source and will not be re-entered into a second system solely for reporting purposes. Data ownership for each exchanged object will be agreed before the interface is enabled.

• ●Curriculum, qualification, and assessment approval authority will remain with CSTARI, DGT, NIMI,

NCVET, or the designated authority. The approved course framework will remain stable during implementation, and future revisions will be made through course versioning and configurable rules.

• ●Participating institutions will have the digital infrastructure, connectivity, and trained users required

for adoption. Where connectivity is constrained, delivery will rely on the offline and low-bandwidth capabilities specified in this report rather than on manual parallel records.

• ●Artificial intelligence capabilities will operate on infrastructure procured or empaneled in accordance

with applicable Government of India IT procurement policy, with the underlying AI model provider abstracted behind an internal AI Gateway (Section 5.2.1) (FIXED - was "Section 5.2") so that no capability in this report is dependent on a single named commercial AI provider.

• ●Approved course content will be indexed into a governed knowledge base to support AI grounding

(Section 4.10); the currency and accuracy of that index against the live course version is a program- level operating responsibility, not solely a one-time build activity.

The table below presents the key stakeholders and the role each performs in the PM-SETU learning ecosystem.

SPV / Anchor Industry Partner

Content / knowledge partner

2. As-Is System LandscapePDF p.7

This section records the digital systems that presently support ITI learning delivery, the role each performs, and the gaps that the PM-SETU Learning Management System is required to close. The description is provided as context for the requirements that f ollow. The business requirements, functional system requirements, architecture, and non-functional requirements set out in the remainder of this report are specified against capability, standards, and interfaces, and are deliberately independent of any single existing platform or product.

2.1 National platforms in current usePDF p.7

Skill India Digital Hub (SIDH): SIDH operates as the national convergence platform for the skilling ecosystem. It maintains learner identity and profile, supports discovery of courses and schemes, provides a learner-facing digital presence, and consolidates participation data for national reporting. SIDH has also introduced AI-assisted course, career-pathway, and skill-gap recommendations, and a conversational assistant, at the national convergence layer, most notably the Skill India Assistant (SIA), an AI-based chatbot deployed on SIDH in partnership with the National Skill Development Corporation, that answers learner queries and guides course and scheme discovery through a widely used messaging channel. SIA is cited in this report as evidence that AI-assisted learner guidance is already an operating pattern within the Government of India skilling ecosystem, not as a capability this platform needs to replicate. In the PM-SETU context, SIDH, including SIA, is significant primarily as an identity, discovery, reporting, and career- intelligence reference layer: it establishes who the learner is and which scheme the learner is associated with, and it operates at national convergence scale rather than at the batch-level teaching record that PM-SETU delivery requires. It is not an instructional delivery system. It does not maintain the batch-level teaching record, the trainer-recorded practical evidence, the assessment attempt history, the module-level progress state, or the intervention record on which day-to-day ITI delivery depends. Pillar 11 (Section 4.11) is accordingly designed to federate with SIDH and its AI-assisted career and recommendation functions rather than duplicate them, consistent with the positioning set out in Section 2.4.

Bharat Skills and NIMI: These national repositories hold curriculum, trade learning material, instructional media, and question banks. They are authoritative sources of approved content, but they are content repositories rather than learning management systems. A learner's consumption of a resource, the sequence in which it was delivered, and the completion arising from it are not recorded against a batch or a course version.

DGT and NCVT MIS: These systems govern trade and institution approval, examination scheduling, result processing, and certificate issuance. They hold the authoritative formal result and the certificate identifier. They operate at examination cycle granularity and do not carry formative assessment, practical evaluation, or in-course progress.

Digi-Locker and the Academic Bank of Credits: These services provide credential custody and academic credit accumulation. They consume credentials issued by an awarding authority; they do not generate them and are not a source of learning activity.

National Career Service (NCS) and NAPS: The National Career Service provides job-matching, skill-gap analysis against job requirements, and career-guidance functions at a national level, while NAPS and NCS together support the post-training stage of the learner lifecycle. Linkage between a completed learning record and an opportunity outcome is presently established manually, where it is established at all, and no system today connects a learner's in-course competency evidence automatically to a job or apprenticeship recommendation.

Bhashini and AI4Bharat: MeitY's Bhashini platform and the AI4Bharat research initiative provide publicly available Indian-language speech, translation, and text infrastructure covering a large number of scheduled Indian languages. These are government-run language infrastructure services, not learning platforms, and

are referenced in this report as the intended mechanism for multilingual and voice-based AI features (Section 4.10, Section 6.6).

Architecture / overviewPDF p.9

Tap to enlarge

2.2 State, institution, and industry systemsPDF p.9

At State level, skill-mission portals maintain admission, attendance, and scheme-reporting data in formats specific to each State. At institution level, the operational teaching record is generally held outside any system of record: attendance registers, spreadsheets for batch and marks data, and messaging groups used to distribute learning material to learners. Biometric or digital attendance capture exists in a few institutions but is rarely linked to the learning record it should inform. A small number of ITIs have procured a learning platform independently; these deployments are isolated and do not exchange records with national systems.

On the industry side, Anchor Industry Partners and original equipment manufacturers frequently operate mature learning platforms of their own, complete with content, assessment, and certification tracks aligned to their equipment and processes; several such platforms internationally have begun to introduce AI-assisted authoring, tutoring, and skills-graph capabilities. These platforms are the authoritative source for the learning they deliver. Separately, publishers and education technology providers supply packaged learning resources without operating any learner-lifecycle system at all. Both categories of partner are essential to PM-SETU delivery, and each requires a different mode of participation.

2.3 Observed gaps in the current landscapePDF p.10

The following observations are drawn from the current-state landscape and determine the capabilities specified in this report.

2.4 Positioning of the PM-SETU LMSPDF p.10

The Learning Management System is positioned as the operational system of record for learning delivery, and only for learning delivery. It establishes and maintains the learner-to-course relationship, the learning and assessment events that occur against it, and the progress, skills, and credential state that results. It does not

assume authority that properly belongs elsewhere: curriculum and qualification approval remains with the designated authority; formal examination and certification remain with the awarding body; credential custody remains with the credential service; national, State, and fiduciary analytics remain with the Analytics Portal; career and labor-market authority remains with the National Career Service and Skill India Digital Hub; and a partner-operated learning platform remains authoritative for the records it owns.

This positioning extends without modification to every artificial intelligence capability specified in this report. An AI capability may generate a draft, a recommendation, an explanation, or a prediction; it does not become a new source of authority. Section 6.8 sets out the AI Governance framework that enforces this positioning across every functional pillar.

Accordingly, every reference in the remainder of this report is expressed in terms of the function required rather than the platform presently performing it. Where the report refers to the approved identity channel, the designated awarding system, the approved credential service, a connected partner platform, or an AI model provider, the specific system fulfilling that role is confirmed during solution design and may change over the life of the program without altering the requirement.

The approved identity channel, designated awarding system, and approved credential service referenced above shall each be confirmed by name before the start of Phase 1 (Annexure F), so that solution design does not proceed against an undetermined dependency.

The distinction between this platform and Skill India Digital Hub is stated concisely here; the full comparison against Skill India Digital Hub, NIMI/Bharat Skills, and commercial platforms is provided at Annexure M.

3. Business RequirementsPDF p.11

The PM-SETU Learning Management System digitizes and governs the end-to-end learning lifecycle of the program through four core business processes. The lifecycle commences with learning setup and enrollment, covering course and batch configuration, learner identity resolution, and course activation, under which the institution establishes the learner-to-course relationship on which every subsequent transaction depends. Once a batch is active, learning delivery processes govern the presentation of the approved content sequence, the conduct of instructor-led and practical activity, the capture of attendance and practical evidence from whichever source owns it, and AI-assisted learning support grounded in the approved course. Attainment is then subjected to assessment and progression, covering LMS-controlled assessment, AI-assisted question authoring and evaluator support, practical evaluation against approved rubrics, the mapping of authoritative external examination results, and advisory adaptive and predictive recommendations. Finally, certification and lifecycle outcomes convert satisfied completion conditions into recognized credentials and, where the relevant integration is enabled, into a skills-based hand-off to on-the-job training, apprenticeship, or employment.

These processes are inherently complex to administer. They span multiple stakeholders - learners, trainers, institution administrators, SPVs and Anchor Industry Partners, industry organizations operating their own

learning platforms, content partners, awarding bodies, and credential services - each with distinct roles, authority, and information needs. They also involve versioned course structures, batch-bound applicability rules, learning events arriving from several systems at different times, assessment results owned in different places, and credentials that only one party is entitled to issue. Managing this lifecycle through manual or fragmented systems produces the conditions observed in Section 2: an incomplete teaching record, progress that cannot be substantiated, duplicate entry wherever an industry partner is involved, and completion claims that outrun the evidence. A dedicated Learning Management System is therefore required to act as the operational system of record - enforcing the approved course version, validating every event against the learner and batch to which it belongs, maintaining complete audit history, applying artificial intelligence only within a governed, advisory boundary, and providing role-appropriate visibility to all stakeholders.

Functional structure

To address these requirements, the Learning Management System is organized into eleven functional pillars, set out in full in Section 4:

• •Pillar 1 - Course Management: the approved catalogue, composition, and version.
• •Pillar 2 - Content Delivery: presentation of approved learning resources across delivery modes.
• •Pillar 3 - Assessment and Evaluation: theoretical and practical measurement, including AI-assisted

assessment support.

• •Pillar 4 - Personalized and Adaptive Learning: bounded, advisory, AI-enabled recommendation of

the next learning action within the approved course.

• •Pillar 5 - Learner Progress Tracking: the consolidated operational record, including AI analytics and

prediction.

• •Pillar 6 - Certification and Badges: recognition of completion.
• •Pillar 7 - Digital Identity and Credential Integration: linkage to national identity and credential

services.

• •Pillar 8 - Student Lifecycle Dashboards and Management: the learner and institution operational

view.

• •Pillar 9 - Interoperability and Integration: governed admission of vendor platforms and content

partners.

• •Pillar 10 - AI Learning Assistance: conversational and generative assistance grounded in approved

course content, for learners and trainers.

• •Pillar 11 - Skills and Career Intelligence: an evidence-based skills profile and advisory career and

employability guidance, federated to national services.

Two cross-cutting frameworks govern every pillar: the Solution Architecture (Section 5), which specifies the shared platform services - including the AI Services block - that pillars draw upon rather than rebuild; and the AI Governance and Responsible AI framework (Section 6.8), which applies a single, consistent set of human- oversight, explainability, auditability, and safety controls to every AI-enabled capability in this report, generalizing the principle first established for Pillar 4.

Role-based access

Access to each pillar, including its AI-enabled capabilities, is scoped to the stakeholder role and entity defined in the User Access Layer (Section 5.4). No AI capability changes a stakeholder's underlying data-access scope; an AI feature exposed to a trainer, for example, operates only within the learners, batches, and institutions already assigned to that trainer.

Architecture / overviewPDF p.13

Tap to enlarge

3.1 LMS access and login journeyPDF p.13

Every user enters through an approved identity route and is resolved to a role and operating scope before any learning record is displayed or changed. Learners are matched to the approved learner identity and enrollment record; official and partner users are matched to the organization and role provisioned for them. Multi-factor authentication is mandatory for administrative, content-approval, assessment-configuration, and AI-configuration roles, consistent with Section 6.2.

3.2 Learner / trainee user journeyPDF p.14

User journeyPDF p.14

Tap to enlarge

• •Landing screen: The learner home

displays the active course, institution, and batch; current progress, attendance, and assessment status; credentials already issued; the next required and next recommended action; and, where enabled, the AI Learning Assistance entry point. It does not expose administrative configuration or another learner's information.

• •Learning delivery:

The learner consumes the approved content sequence for the active course version and may use AI-generated summaries, flashcards, and practice questions grounded in that content as optional study aids, clearly distinguished from the mandatory sequence (Pillar 10).

• •Doubt resolution and study support:

Where enabled by the institution, the learner may ask the AI Course Q&A Assistant a question about the current course; the assistant answers only from approved course content, with a visible source reference, and states explicitly when it cannot answer rather than generating an unsupported response (Pillar 10).

• •Assessment and adaptive learning:

The learner attempts the LMS- controlled assessment or receives a practical evaluation. Where the result identifies a permitted learning gap, the system may recommend remedial content, practice, or a personalized study plan. The recommendation is advisory and cannot change a mandatory module, official result, pass threshold, or certification requirement (Pillar 4).). Where the course requires it, the learner's journey begins with a baseline assessment at course activation and concludes with a matched endline assessment at completion, with the resulting learning- gain measure presented alongside standard progress.

• •Cohort engagement: Where the institution has enabled it and the learner has opted in, the learner sees

their standing on a batch-scoped engagement ranking based on recorded completion metrics; participation is off by default, has no effect on any mandatory requirement, assessment, or credential, and can be withdrawn by the learner at any time (Pillar 8).

• •Progress and skills: The learner views consolidated progress and, where the course is mapped to

competencies, an evidence-based skills profile and any advisory career or job-matching guidance federated from the National Career Service and Skill India Digital Hub (Pillar 11).

• •Credentials: The learner views or downloads their own progress statement and issued credentials once

confirmed by the issuing authority.

3.3 Trainer / faculty user journeyPDF p.15

User journeyPDF p.15

Tap to enlarge

• •Teaching workspace:

The trainer accesses assigned courses, batches, and learners, records attendance and practical evidence, and administers assessment within the assigned scope.

• •AI-assisted preparation:

Where enabled, the trainer may use the Trainer Copilot to draft session plans and supporting material from approved course content, and the Trainer Content Q&A Assistant to query course sequencing and delivery guidance; every draft requires trainer review before use (Pillar 10).

• •Assessment authoring and evaluation

support: The trainer or assessor may use AI-assisted question and distractor generation, difficulty prediction, and rubric-assisted or descriptive-answer scoring support; the trainer or assessor confirms, edits, or rejects every AI- generated item or suggested score before it is used or recorded (Pillar 3).

• •Learner progress and intervention: The

trainer reviews module completion, attendance, assessment status, and advisory learning-gap and at-risk indicators for assigned learners, and records the remedial action, follow-up, and outcome where intervention is required. An adaptive or predictive recommendation may be accepted, deferred, or replaced by an approved alternative, with action retained in history (Pillar 4, Pillar 5). Where a learner's difficulty appears personal or circumstantial rather than academic, the trainer may record a counselling referral in the same intervention history, distinct from an academic remedial action.

• •Natural-language data access: Where enabled, the trainer may query assigned learner and batch data

using plain-language questions, scoped strictly to the trainer's existing assignment (Pillar 10).

3.4 ITI administrator user journeyPDF p.16

• •Course and batch operations: The

User journeyPDF p.16

Tap to enlarge

institution administrator activates approved course versions, creates batches, and assigns trainers and learners within the institution's authorized scope.

• •Exception and data-quality management:

The administrator resolves institution-level data-quality exceptions, including AI-flagged anomalies, duplicates, and reconciliation issues, all of which are presented as advisory findings for administrator review, not automatic corrections (Pillar 8, Platform Layer Admin AI service).

• •Institution analytics:

The administrator views cohort-level and course-effectiveness analytics and any prescriptive institutional recommendations, which remain advisory and require an explicit administrator’s decision to act upon (Pillar 5, Pillar 8).

• •Reporting: The administrator may

generate operational reports using natural-language queries and AI- generated summaries, reviewed before circulation (Platform Layer Admin AI service).

3.5 SPV / Anchor Industry Partner user journeyPDF p.17

User journeyPDF p.17

Tap to enlarge

• •Cluster delivery coordination:

The SPV or Anchor Industry Partner uses the system to operationalize the approved academic and industry-delivery model across its Hub-and-Spoke cluster, while preserving the operating autonomy and source ownership of the institutions and industry systems that create the underlying learning events.

• •Industry and course-mapping

inputs: The SPV/AIP workspace presents cluster-level industry and course-mapping inputs for review.

• •Reconciliation status:

The SPV/AIP workspace presents the reconciliation status of learning records across the cluster's institutions.

• •Cluster-level AI analytics: The

SPV/AIP views cluster-level AI analytics, sourced from the same advisory Pillar 5 predictive service used at institution level, scoped to the cluster's own institutions.

3.6 Industry partner with existing LMS user journeyPDF p.18

User journeyPDF p.18

Tap to enlarge

• •Interface registration: An industry

partner that already operates a learning platform registers the interface, scope, and data contract through the integration gateway (Pillar 9).

• •Activity exchange: The partner

exchanges the approved subset of learner activity module completion, assessment results, practical and mentor evaluations, and credential references without re-entering that activity in PM-SETU.

• •AI-sourced record reconciliation:

Where the partner's own platform provides AI-generated learning content or assessment results, those records are received and reconciled as externally sourced records.

• •AI boundary: PM-SETU does not re- process or re-score partner-sourced content or results through its own AI services.

3.7 Content / knowledge partner user journeyPDF p.19

User journeyPDF p.19

Tap to enlarge

• •Content contribution: A content-only

partner contributes approved learning resources with rights and metadata, without operating a learner-lifecycle system.

• •Academic review: Content submitted by the partner is validated by the academic reviewer against the course mapping.
• •AI grounding eligibility: Only after that

validation, and only once the content has been published under an active course version, does it become eligible for indexing into the governed knowledge base described in Section 4.10 and Section 5.2 for AI grounding purposes.

• •AI boundary:

Unpublished or unmapped content is never indexed or surfaced through any AI capability.

3.8 Academic reviewer user journeyPDF p.20

User journeyPDF p.20

Tap to enlarge

• ●Content and mapping review: The

academic reviewer reviews submitted content against the course mapping (Pillar

• 2)and reviews partner course or content mappings (Pillar 9), approving each submission or returning it for correction.
• ●AI-grounding eligibility gate:

The reviewer's approval is the trigger that makes published content eligible for indexing into the governed knowledge base used for AI grounding (Pillar 10); unapproved or unmapped content is never indexed or surfaced through any AI capability.

• ●Scope of authority: The academic

reviewer's access is scoped strictly to the review and approval action on submitted content and mappings, not general content authoring or connector configuration, consistent with Annexure A.

User journeyPDF p.20

Tap to enlarge

3.9 Model owner user journeyPDF p.20

• ●AI Governance console: The model

owner accesses the AI Governance console, including model and rule version history and the disparity-review dashboard, across every AI-enabled pillar.

• ●Model and rule version approval: The

model owner approves a new or updated AI model or rule configuration before it applies prospectively, consistent with the Model Registry requirement in Section 6.8.2.

• ●Predictive-indicator suspension: The

model owner may suspend a Pillar 5 predictive indicator found to exhibit a quality or disparity issue, pending review.

• ●Periodic disparity review: The model

owner conducts the periodic bias and disparity review across language, region, gender, and institution type required under Section 6.8.2, and reports findings to the AI Governance function specified in Annexure

E.

3.10 Integration administrator user journeyPDF p.21

User journeyPDF p.21

Tap to enlarge

• ●Partner integration console: The

integration administrator accesses the partner integration console showing connector status, last synchronization, and outstanding reconciliation exceptions by partner (Pillar 9).

• ●Connector configuration:

The integration administrator configures a new connector's credentials, scope, and identifier mapping during partner onboarding, and monitors ongoing connector health.

• ●Exception resolution: The integration

administrator resolves mapping and reconciliation exceptions arising across course mapping (Pillar 1), content ingest (Pillar 2), and connector-level integration points (Pillar 9).

4. Functional System RequirementsPDF p.22

This section sets out the functional requirements of the eleven pillars of the PM-SETU Learning Management System. The pillars follow the learner lifecycle: Course Management establishes what is to be learned; Content Delivery presents the approved learning resources; Assessment and Evaluation measures attainment, with AI-assisted authoring and evaluation support; Personalized and Adaptive Learning determines the next recommended action within the approved course; Learner Progress Tracking maintains the running record and advisory predictive indicators; Certification and Badges recognizes completion; Digital Identity and Credential Integration links the learner to authoritative identity and credential services; Student Lifecycle Dashboards and Management presents the resulting journey; Interoperability and Integration admits external learning platforms and content sources under a governed contract; AI Learning Assistance provides grounded conversational and generative support to learners and trainers; and Skills and Career Intelligence extends course-level competency evidence into a durable skills profile and advisory employability guidance.

Each pillar is specified in a consistent form. The description defines the business purpose; Functional Capabilities state what the system shall provide; and the Acceptance Criteria, Validation Rules, and Exception Management sections define the conditions under which the capability is considered fit for use. Where a pillar includes artificial intelligence, its AI Behavior, Human Control, and Governance are stated explicitly and are subject, without exception, to the cross-cutting AI Governance framework in Section 6.8. Requirements are expressed against capability and standard rather than against any product.

4.1 Pillar 1: Course ManagementPDF p.22

The system shall establish and maintain the approved learning structure presented to a PM-SETU learner. It should maintain the course and qualification catalogue, the version applicable to each batch, the relationship between national core content, approved industry modules, practical activity, on-the-job training, and other approved components, and the metadata required to identify a course consistently across institutions and partner systems.

The component is operated by authorized curriculum and institution administrators and is consumed by learners and trainers through the assigned course. Formal approval by CSTARI, DGT, NIMI, NCVET, or another designated authority is not replaced by the system; the platform records and enforces the approved version and its supporting evidence. The course version and its associated competency map are also the reference structure used by Pillar 4, Pillar 10, and Pillar 11 to bound their AI-enabled capabilities.

BPMN / process flowPDF p.23

Tap to enlarge

Functional CapabilitiesPDF p.23

Catalogue and qualification mapping

• •Maintain a searchable course and qualification catalogue with trade, occupation, NSQF level, QP-NOS

reference, credit, duration, delivery mode, and certification metadata.

• •Map each course to the approved national qualification record maintained by the designated authority.
• •Record the institutions, clusters, and States for which a course is applicable.
• •Support search and filter by trade, level, delivery mode, language, and institution, and expose the

catalogue to the AI-Powered Course Catalogue Search capability specified in Pillar 10.

Course composition and creation

• •Create modular course structures comprising national core, approved industry modules, practical

activities, on-the-job training, and other approved components.

• •Define prerequisites, sequencing rules, and the conditions under which a module unlocks.
• •Associate each course with its content set, assessment plan, completion criteria, competency map, and

applicable credential type.

• •Capture the approval evidence, effective date, and authority for the composition.

Versioning and applicability

• •Maintain version, effective date, status, and batch applicability for every course.
• •Retain a live batch on the course version on which it commenced unless an approved transition is

recorded.

• •Preserve every superseded version and the batches that ran against it.
• •Prevent in-place editing of a published version; a material change creates a new version.

Batch and cohort management

• •Create and manage batches and cohorts by trade, shift, and institution.
• •Assign trainers and learners to a batch within the institution's authorized scope.
• •Capture the session and lesson delivery calendar applicable to the batch.
• •Maintain batch status through planning, active, completed, and closed states.

Partner and institution course mapping

• •Map partner- or institution-owned courses to the PM-SETU course identifier and taxonomy.
• •Preserve the partner's internal identifier alongside the canonical identifier.
• •Hold an unmapped or ambiguous partner course as an exception rather than publishing it.
• •Support reconciliation reporting between partner and PM-SETU course catalogues.

Acceptance CriteriaPDF p.24

• •Every learner and batch is linked to one active course version, and that version remains traceable after

later revisions.

• •Course metadata displayed to the learner agrees with the approved course record and the institution

and batch assignment.

• •A partner course can be reconciled to both the partner identifier and the PM-SETU course identifier.
• •A course version cannot be published while mandatory approval evidence or metadata is incomplete.
• •Batch creation, trainer assignment, and learner assignment are completed digitally within the institution's

authorized scope.

Validation RulesPDF p.24

• •A published course version cannot be edited in place; a material change creates a new version.
• •Mandatory approval evidence, ownership, and effective date are required before publication.
• •A course cannot be assigned to a batch outside its approved applicability or effective period.
• •Formal examination or certification rules held in an external authoritative system are referenced rather

than overwritten.

• •A learner cannot be assigned to more than one active batch of the same course version.

Exception ManagementPDF p.25

4.2 Pillar 2: Content DeliveryPDF p.25

The system shall present the approved teaching-learning material associated with the course version and support the blended learning model required for PM-SETU. It shall cover instructor-led and self-learning resources, multimedia, digital learning assets, and practical support material, while retaining source, ownership, version, rights, language, and accessibility information for each resource. Published content is the exclusive source corpus indexed by the knowledge base used for AI grounding in Pillar 3, Pillar 4, and Pillar 10; unpublished or unmapped content is never indexed.

Functional CapabilitiesPDF p.25

Resource presentation and blended delivery

• •Present instructor-led, self-paced, multimedia, and practical support resources in the sequence defined

by the course version.

• •Support online, offline, and low-bandwidth delivery modes, with deferred synchronization of offline

activity.

• •Retain source, ownership, version, rights, language, and accessibility metadata for every resource.
• •Present an accessible alternative in place of, rather than in addition to, an inaccessible resource,

consistent with Section 6.6.

Content ingests and partner content

• •Ingest standards-based content packages (SCORM, xAPI, cmi5, QTI 3.0) and support remote launch of

partner-hosted content under LTI 1.3.

• •Validate a submitted package against the agreed conformance standard before publication.
• •Route an unmapped or rights-incomplete

BPMN / process flowPDF p.26

Tap to enlarge

submission to academic review rather than publishing it.

Activity capture and reliability

• •Record reliable content-activity events - access,

progress, and completion - against the learner, course version, and batch.

• •Reconcile partner-delivered content activity

received through the integration gateway with the PM-SETU learner record.

• •Preserve historical completion evidence when a

resource's rights expire or the resource is withdrawn.

Acceptance CriteriaPDF p.26

• •The content presented to a learner matches the

approved sequence of the active course version.

• •Every content-activity event is attributable to a

learner, course version, batch, and source.

• •A resource whose rights have expired is

withdrawn from new delivery while historical completion remains traceable.

• •Offline activity synchronizes without creating a

duplicate event.

Validation RulesPDF p.26

• •A resource whose validity period has ended

cannot be assigned to a new activity.

• •Deferred offline activity must carry the original

activity timestamp and cannot overwrite a later validated event.

• •Content is not indexed for AI grounding until it is

published under an active course version and validated by academic review.

Exception ManagementPDF p.27

4.3 Pillar 3: Assessment and EvaluationPDF p.27

The system shall support the measurement of theoretical knowledge and practical skill during the learning period. It shall provide the assessment workflow for formative checks, internal assessments, and practical evaluation, while preserving the authority of the designated examination or awarding system for formal results, and shall provide governed AI assistance for question authoring, adaptive practice, and evaluator support without transferring assessment authority to the AI.

The component is used by learners, trainers, and authorized institution users. Industry mentors may provide evaluation input for on-the-job training or practical activity only where the approved assessment plan assigns that role. The distinction between an LMS-controlled result and an externally authoritative result is enforced throughout: the system computes the former and maps the latter, and no AI capability in this pillar is permitted to compute, override, or recalculate an externally authoritative result.

BPMN / process flowPDF p.28

Tap to enlarge

Functional Capabilities - Core AssessmentPDF p.29

Assessment planning

• •Maintain assessment plans against the course version, including assessment type, applicable module

or competency, attempt rule, threshold, and weighting.

• •Record for each assessment whether the result is LMS-controlled or externally authoritative.
• •Define the role permitted to record each practical or mentor evaluation.
• •Schedule assessments within the learner journey and the batch calendar.
• •Support a baseline assessment recorded at course activation and a matched endline assessment

recorded at course completion, both LMS-controlled, using the same question-bank and scoring mechanism as any other assessment type in this pillar.

• •Compute a learning-gain measure as the difference between a learner's endline and baseline result on

matched competencies, presented to the learner, trainer, and institution as a recorded outcome, not an AI-generated estimate.

Assessment content and rubrics

• •Maintain question-bank and assessment-content references with competency, difficulty, and version

metadata.

• •Support multiple assessment formats, including objective, descriptive, practical, and viva components.
• •Maintain practical rubrics with defined criteria and outcome levels for trainer or mentor evaluation.
• •Support import of externally authored assessment content in the agreed interchange format.

Attempt management and scoring

• •Present the assessment to the learner, record the attempt, and preserve the evidence required for

review.

• •Calculate LMS-controlled scores in accordance with the configured rule.
• •Record trainer-evaluated practical outcomes with the evaluating user, rubric version, and timestamp.
• •Apply the configured resume or restart rule where a session is interrupted and record the interruption.
• •Enforce the configured attempt rule, including permitted attempts and any cooling -off period.

Authoritative external results

• •Import formal examination results or status from the authoritative examination or awarding system.
• •Match the imported result to learner, course version, and assessment before accepting it.
• •Present the mapped result without recalculating, moderating, or overriding it, by any component, human

or AI.

• •Expose the received status and its source and freshness to the learner and trainer.

Downstream exposure

• •Expose assessment status to the progress, adaptive-learning, and AI analytics components in read-only

form.

• •Record a re-evaluation or correction as a new controlled event with reason and audit trail.
• •Emit validated assessment events to the learner record and the learning record store.

Functional Capabilities - AI-Assisted Assessment SupportPDF p.29

This capability provides generative and predictive AI assistance to authors, trainers, and assessors across question authoring, adaptive practice, and evaluator support. It does not alter the assessment authority

defined above: every AI-generated or AI-suggested output in this subsection is Advisory or Human-approved, never Authoritative, under the AI output classification defined in Section 6.8.

AI-assisted question and item authoring

• •Generate draft objective, descriptive, and practical questions and plausible distractors aligned to a

defined learning outcome or competency, grounded in the approved course content indexed under Pillar

10.

• •Require author or assessor review and explicit approval before a generated item enters a live question

bank; a generated item is never published without this confirmation.

• •Predict the difficulty level of a bank item or a newly generated item using an item-response model

calibrated on historical response data, presented as an estimate for author review, not a fixed classification.

• •Assemble a test from the approved question bank to match a defined assessment blueprint, subject to

author confirmation before use.

Adaptive assessment (formative use)

• •Where a course administrator explicitly enables it for a formative or practice assessment, adjust question

difficulty in real time based on the learner's responses within that session.

• •Prohibit adaptive difficulty adjustment on any assessment configured as the basis for a certification,

pass/fail, or completion decision, unless the course administrator has explicitly and separately authorized it for that use, in which case the underlying pass rule itself is never altered by the adaptive layer.

AI-assisted evaluator support

• •For descriptive or essay-type answers, generate a suggested score against the defined rubric together

with a stated rationale; display the suggestion alongside, and visually distinct from, the evaluator's own scoring control.

• •Treat the evaluator's confirmed score as the sole recorded result; an AI-suggested score that is not

explicitly confirmed by the evaluator is never recorded as the assessment outcome.

• •Apply defined practical-evaluation rubric to support consistent trainer scoring of practical work,

presented as a scoring aid; the trainer's recorded evaluation remains authoritative.

• •Generate explanatory feedback on assessment performance for learner review, grounded in the specific

items attempted and the applicable rubric or answer key.

• •Analyze recorded response patterns to identify recurring misconceptions and surface assessment-level

analytics against defined learning outcomes for author and institution review.

Assessment-integrity signals

• •Flag a submitted written answer with a likelihood-of-plagiarism signal for evaluator review; the signal

alone does not constitute a finding and does not trigger any automatic penalty.

• •Flag an answer with an indication that it may have been produced by a generative AI tool, presented

strictly as an advisory signal for evaluator review; this indication shall not, by itself, be treated as proof of misconduct, shall never independently trigger a penalty or disciplinary action, and any action taken on the basis of such a flag remains a human evaluator decision, given documented elevated false- positive rates of such detection tools against non-native English writers.

Acceptance CriteriaPDF p.30

• •Every recorded result identifies the learner, course version, assessment, attempt, evaluator or source,

and timestamp.

• •A learner sees the assessment status and feedback permitted by the assessment policy and cannot see

another learner's record.

• •An externally authoritative result is presented without being recalculated by the system or by any AI

capability.

• •Concurrent submission of an attempt does not create a duplicate result.
• •A practical evaluation can be recorded only by the role named in the applicable assessment plan.
• •No AI-generated question enters a live assessment without a recorded author or assessor approval

action.

• •No AI-suggested score becomes the recorded assessment result without an explicit evaluator

confirmation captured in the audit trail.

• •A learner's endline result is only computed against a baseline result recorded for the same learner and

course version; an endline attempted with no matching baseline is flagged rather than silently scored.

Validation RulesPDF p.31

• •Only the role defined in the assessment plan may record a practical evaluation.
• •A completed attempt is immutable; a re-evaluation or correction is recorded as a new controlled event

with reason and audit trail.

• •The system must not record course completion whereas the configured rule requires an authoritative

result that has not been received.

• •An attempt beyond the configured attempt limit shall not be created.
• •An imported result must reference a valid learner, course version, and assessment before it is accepted.
• •Adaptive difficulty adjustment shall not modify the pass threshold or certification requirement of the

assessment it is applied to.

• •AI-generated content that fails an automated quality or grounding check is rejected and is not presented

to an author for approval.

Exception ManagementPDF p.31

4.4 Pillar 4: Personalized and Adaptive LearningPDF p.32

The system shall use the learner's recorded progress, assessment performance, completed competencies, and other approved learning signals to recommend the next learning action and to generate a personalized study plan. This capability implements the PM-SETU intent for personalized pathways and intelligent tutoring support, but should not alter the approved curriculum, an official result, or a certification threshold. The learner remains on the approved course; the adaptive layer changes the sequence, pace, or recommendation of supplementary learning within the boundaries configured for that course.

The recommendation is presented to the learner and trainer as an advisory next action. A trainer may accept, defer, or replace the recommendation with an approved alternative, and the action taken is recorded. Every recommendation shall be explainable by reference to the recorded signal that produced it. This pillar is the origin of the AI Governance principles - advisory-only output, explainability, human override, and prospective- only versioning - that Section 6.8 generalizes across every AI capability in this report.

BPMN / process flowPDF p.33

Tap to enlarge

Functional CapabilitiesPDF p.33

Recommendation generation

• •Generate recommended next content, practice, remedial activity,

or revision based on the learner's current course state and configured pedagogic rules.

• •Identify individual skill gaps by reference to the competency map

of the active course version.

• •Recommend both remedial and advanced content according to

demonstrated attainment.

• •Use assessment and progress signals only from records within

the learner's active course.

• •Generate a personalized, time-boxed study or revision plan from

the learner's recorded progress and upcoming assessment schedule, presented as an editable suggestion rather than a fixed obligation.

Difficulty and pace adaptation

• •Adjust the presented difficulty of recommended practice content,

and the suggested pace of a personalized study plan, to the learner's demonstrated ability within the active course.

• •Apply difficulty and pace adaptation only to supplementary and

formative content; never to a mandatory module, assessment threshold, or completion condition.

Boundaries and configuration

• •Permit course administrators to define which modules may be

reordered or supplemented and which are fixed prerequisites.

• •Prevent any recommendation from waiving a mandatory module,

practical activity, assessment, or certification requirement.

• •Prevent the adaptive layer from writing a score, pass or fail

status, completion state, or credential.

• •Version rule and model configuration and apply changes

prospectively only.

Explainability and human control

• •Identify the competency or recorded signal that caused a

recommendation and present a concise reason to the learner and trainer.

• •Distinguish adaptive output visually and structurally from

mandatory requirements and recorded outcomes.

• •Permit a trainer to accept, defer, or override an advisory

recommendation with a reason, without changing the underlying recorded performance.

• •Record recommendation, acceptance, or override and

subsequent outcome so that effectiveness can be evaluated.

Data protection in recommendation

• •Use only the learning signals defined in the approved model configuration for the active course.
• •Do not combine unrelated personal data solely for the purpose of generating a recommendation.
• •Retain the input signal set associated with each recommendation for audit and review.
• •Support periodic review of recommendation quality and of any observed disparity in outcomes across

learner groups.

Acceptance CriteriaPDF p.34

• •Every recommendation is traceable to the active course version and the learner signals used to generate

it.

• •The learner can continue to see the approved learning path even when the recommendation service is

unavailable.

• •Adaptive output is visibly distinguished from formal requirements and recorded outcomes.
• •A trainer override is recorded with reason and does not alter the recorded performance that prompted

the recommendation.

• •Model and rule versions in force at the time of each recommendation are recoverable.

Validation RulesPDF p.34

• •No recommendation may waive a mandatory module, practical activity, assessment, or certification

requirement.

• •A recommendation cannot write a score, pass or fail status, completion state, or credential directly.
• •Model and rule changes are versioned and applied prospectively to recommendations generated after

activation.

• •Where a recommendation conflicts with a fixed prerequisite or trainer restriction, the fixed course rule

prevails.

• •Recommendation output shall not be presented to any stakeholder as an assessment result.

Exception ManagementPDF p.34

4.5 Pillar 5: Learner Progress TrackingPDF p.35

BPMN / process flowPDF p.35

Tap to enlarge

The system shall maintain the operational record of a learner's advancement through the assigned course and support timely intervention by the learner's institution. It shall consolidate module completion, attendance, practical evidence, assessment status, interventions, and completion criteria into a single course-level record while retaining the source of each event, and shall generate firm, advisory, explainable AI predictive indicators to support earlier institutional intervention.

The progress record is the basis for the learner dashboard and for operational hand-off to the Analytics Portal. Program-wide aggregation and threshold analytics are performed in that portal and are not duplicated here.

This pillar consolidates the learner's progress record, but does not itself perform registration, baseline assessment, or career-path discovery - each of these sits at a different, earlier or later stage of the same learner lifecycle this pillar tracks. Registration and enrollment are resolved, and every staff and partner user is provisioned, before any learning activity is created (Pillar 7). Where the course requires it, a baseline assessment is recorded at course activation and a matched endline assessment at course completion (Pillar 3), with the resulting learning-gain measure available alongside the progress this pillar maintains. Advisory career path discovery, job-matching, and labor- market-informed guidance, built from the same recorded evidence this pillar consolidates, are provided in Pillar 11.

Functional Capabilities - Progress TrackingPDF p.35

Progress state maintenance

• •Maintain module and activity status, attendance, practical

evidence, assessment status, and intervention history for each learner and course version.

• •Calculate course completion from the approved completion rule

and display the components still outstanding.

• •Present competency or skill-gap status where the course is

mapped to competencies and the evidence required for attainment is available.

• •Maintain the state separately for each active course where a

learner is enrolled in more than one.

Event validation and lineage

• •Update progress only from events validated against the learner,

course version, batch, and activity.

• •Retain the source identifier, recording user or system, and

freshness timestamp for every event.

• •Suppress and retain duplicate events for reconciliation rather

than applying them.

• •Record a correction as a subsequent event with source and reason, preserving the original evidence.

Intervention

• •Permit authorized trainers and institution users to record an intervention, the action agreed, and the

follow-up date, preserving the history of prior interventions.

• •Support recording a referral to institution-level psychosocial or personal counselling support as a distinct

intervention type, separate from academic remediation, with its own follow-up date and outcome; the platform records the referral and its outcome and does not itself provide counselling.

• •Record dropout or discontinuation with the recorded reason and retain the learner's history.

Publication to monitoring

• •Publish validated learning events and reconciliation status to the Analytics Portal or program monitoring

layer.

• •Expose the source and freshness of each published event.
• •Prevent any analytics or monitoring consumer from editing the progress record.
• •Support scheduled and event-driven publication according to the agreed interface.

Functional Capabilities - AI Analytics and PredictionPDF p.36

This capability replaces the general notion of “predictive or risk indicators, where used” with a firm, specified requirement. Every output in this subsection is Advisory: it informs a trainer or institution decision and never itself changes progress, completion, or credential state.

Predictive indicators

• •Generate an advisory dropout or attrition risk indicator for each learner from validated attendance,

progress, and engagement signals.

• •Generate advisory completion-likelihood, assessment-performance, and engagement-level indicators

from the same validated signal set.

• •Explain each indicator with reference to the specific signals that produced it, in the same manner as the

Pillar 4 recommendation explainability requirement.

• •Suppress a prediction rather than infer one where the learner's signal history is insufficient or stale.

At-risk identification and intervention support

• •Flag learners who require attention based on configured operational conditions, informed by the

predictive indicators above, such as low attendance, stalled progress, or repeated assessment difficulty.

• •Present every predictive or risk indicator as advisory rather than determinative, on every screen where

it appears.

• •Group learners with similar identified gaps to support targeted remedial sessions, as an advisory

grouping only.

Institution and cluster analytics

• •Analyze cohort- and behavior-pattern trends across batches, institutions, or clusters for institution and

SPV review, presented through Pillar 8.

• •Analyze which courses or content are associated with better learner outcomes, for author and

administrator review.

• •Generate prescriptive institutional recommendations from predicted cohort outcomes; such

recommendations are advisory only and are never executed automatically.

Governance of predictive analytics

• •Apply periodic disparity review of prediction outcomes across language, region, gender, and institution

type, consistent with Section 6.8.

• •Retain the input signal set associated with each prediction for audit and review, consistent with the data-

minimization principle already established in Pillar 4.

Acceptance CriteriaPDF p.37

• •The learner and trainer see the same underlying recorded progress for the active course, subject to role-

appropriate detail.

• •Course completion reconciles with the mandatory learning and assessment requirements of the active

course version.

• •Every externally supplied learning event retains its source identifier and timestamp.
• •An outstanding component is visible to the learner as pending, with the specific dependency named.
• •Intervention history is retained in full and is attributable to the recording user.
• •Every predictive indicator is explainable by reference to the signals used and is visually distinguished

from a recorded outcome.

• •No predictive indicator, by itself, changes a learner's progress, completion, or credential state.

Validation RulesPDF p.37

• •Progress is updated only from validated events linked to the learner, course version, and batch.
• •A corrected event does not delete the original audit evidence; the correction is recorded with source and

reason.

• •Analytics and monitoring consumers, including AI prediction services, cannot edit the progress record.
• •Completion shall not be completed while the required authoritative result is absent.
• •An event whose timestamp precedes the batch activation date shall not be applied without review.
• •A prescriptive institutional recommendation shall not trigger an automated action without an explicit

administrator’s decision.

Exception ManagementPDF p.37

4.6 Pillar 6: Certification and BadgesPDF p.38

The system shall recognize learner completion and present the status of credentials arising from the learning journey. It shall support completion certificates, digital badges, and stackable micro-credentials where the program permits them, and shall map formal credentials issued by the designated awarding body. The system

shall not create a competing formal certificate where an external awarding system is authoritative, and no AI capability in this report is permitted to issue, alter, or influence a credential decision.

BPMN / process flowPDF p.39

Tap to enlarge

The learner sees credential status in the same workspace as learning progress, with a verifiable reference or link to the authoritative credential service where one is available.

Functional CapabilitiesPDF p.39

Completion evaluation

• •Evaluate the configured completion criteria before creating any

completion certificate or badge.

• •Display the specific conditions outstanding where completion

criteria are not satisfied.

• •Distinguish course progress from course completion in every

learner-facing presentation.

• •Re-evaluate completion automatically when an outstanding

dependency is subsequently satisfied.

Credential issuance and mapping

• •Maintain credential type, course version, learner, issue date,

issuing authority, status, and verification reference.

• •Map issued certificates to the applicable NSQF qualification level

where the credential is so aligned.

• •Support stackable module badges and micro-credentials for

partial completion where approved for the course.

• •Receive or map formal certification status from the designated

awarding system and present it without changing the authoritative record.

Verification and standards

• •Generate a learner-facing verification reference or machine-

readable code where the credential service and program policy permit it.

• •Issue digital badges and micro-credentials in an open, verifiable

format aligned to Open Badges 3.0 or W3C Verifiable Credentials.

• •Expose the issuing authority and credential type on every

credential presented.

• •Support publication of the credential reference to an approved

credential service.

Lifecycle and revocation

• •Reflect a revoked or corrected credential status without deleting

the historical audit event.

• •Publish credential events to the learner progress record and

downstream consumers after successful validation.

• •Retain the completion evidence that supported issuance for the applicable retention period.
• •Prevent presentation of a credential as issued before the issuing service confirms issue.

Acceptance CriteriaPDF p.40

• •A credential is not shown as issued until the issuing or authoritative service confirms issue.
• •Every credential presented to the learner identifies the issuing authority and credential type.
• •A revoked or corrected credential status is reflected without deleting the historical audit event.
• •A badge or micro-credential is verifiable by a party outside the platform.
• •Course progress is never presented as a complete qualification.

Validation RulesPDF p.40

• •Credential issuance requires all completion conditions defined for that credential type.
• •The system must not synthesize a formal certificate identifier where the awarding body owns that

identifier.

• •External credential references are validated for learners and courses or qualification match before

display.

• •A micro-credential shall not be represented as a formal qualification unless the approving authority has

defined it as such.

• •A credential cannot be issued against a course version other than the one the learner completed.

Exception ManagementPDF p.40

4.7 Pillar 7: Digital Identity and Credential IntegrationPDF p.40

The system shall link the learner record to approved national identity, profile, and credential services so that a learner is not required to maintain a separate identity for each learning function. It shall provide identity resolution, role and entity mapping, consent and data-minimization controls, and governed connectors to the approved identity, credential, and academic credit services.

This component links and authorizes; it does not replicate. Identity source data, credential documents, and external credential ledgers remain in their designated systems, and the platform stores only the identifiers, status, and references required to render the learner journey and prove the source of the information.

BPMN / process flowPDF p.41

Tap to enlarge

Functional CapabilitiesPDF p.41

Identity resolution and role mapping

• •Resolve every learner to the approved learner identity

and enrollment record before any learning activity is created.

• •Resolve every staff and partner user to a provisioned

role, institution, and organization before granting access.

• •Maintain the mapping between the PM-SETU identifier

and the external identity reference without exposing unrelated identity attributes.

Consent and data minimization

• •Record consent status for the specific data elements

shared with each connected external service.

• •Limit the profile attributes exchanged with any external

service to the agreed data contract.

• •Apply the age-appropriate consent pathway required

under applicable data-protection law where the learner is a minor, including verifiable parental consent where required.

Credential and academic-credit linkage

• •Publish the reference to an issued credential to the

approved credential service, such as DigiLocker, where enabled.

• •Post academic-credit references to the Academic Bank

of Credits where the course and program are so configured.

• •Retrieve and present the status of a credential or credit

posting without duplicating the underlying document.

Acceptance CriteriaPDF p.41

• •A user cannot access any learning record before identity

resolution succeeds.

• •Consent status is recorded for every data element

shared with an external service.

• •A credential or credit reference presented to the learner

matches the confirmed status of the issuing service.

Validation RulesPDF p.41

• •Identity resolution failure prevents access rather than

granting a default or partial scope.

• •Data exchanged with an external service is limited to the

fields defined in the agreed contract.

• •A minor's data is processed only within the consent pathway required by applicable law.

Exception ManagementPDF p.42

4.8 Pillar 8: Student Lifecycle Dashboards and ManagementPDF p.42

The system shall provide the learner-facing and institution-facing consolidation of the learning record. It shall present the learner's active course, next action, progress, assessment status, skills, credentials, and lifecycle milestones from enrollment through on-the-job training or apprenticeship to completion. The same operational record shall be available to the learner's trainer and institution at the level required to support delivery and intervention.

This component is intentionally limited to the learner and institution operational journey. State, national, fiduciary, and cross-system dashboards are delivered through the Analytics Portal and are not recreated here. Administrative AI outputs - natural-language reporting, report summaries, anomaly and duplicate detection, and support-ticket classification - are processed by the shared Admin AI service in the Platform Layer (Section 5.2) and are presented to institution and program administrators through this pillar; the underlying processing does not reside here.

BPMN / process flowPDF p.43

Tap to enlarge

Functional CapabilitiesPDF p.43

Learner workspace

• •Present a learner home showing active course and batch, current

module, next required action, next recommended action, progress, upcoming assessment, skills profile, and credential status.

• •Provide a chronological lifecycle view covering enrollment, course

activation, learning milestones, assessments, interventions, completion, credential status, and opportunity hand-off.

• •Present only those opportunities and support actions relevant to the

learner's approved program context and integration scope.

• •Allow the learner to view or download the learner's own progress

statement and credential references where permitted.

Trainer and institution views

• •Provide the trainer with a batch-level operational view of learner

progress, pending evaluations, and interventions.

• •Provide the institution with a view of active and upcoming courses and

batches, trainer and learner assignment status, delivery and assessment readiness, and unresolved exceptions, including AI- flagged data-quality exceptions.

• •Restrict every view to the assigned institution, courses, batches, and

learners.

• •Support export of institution-scoped operational reports, including AI-

generated report summaries reviewed before circulation, in standard formats.

Attrition and lifecycle indicators

• •Track discontinuation and attrition with the recorded reason and retain

the learner's history.

• •Present lifecycle stage and elapsed duration against the planned batch

calendar.

• •Surface learners whose lifecycle has stalled at a specific stage for

institution follow-up, informed by the Pillar 5 predictive indicators.

• •Distinguish advisory risk and predictive indicators from recorded

outcomes in every presentation.

Data quality and transparency

• •Expose data-quality and missing-dependency status in plain language

rather than displaying an incomplete record as complete.

• •Derive every displayed status from an operational record or an

explicitly stored workflow state; no AI-generated summary is treated as a new source of truth.

• •Show the freshness of externally sourced status on the screen where

it appears.

• •Retain the last confirmed state where a source component is

unavailable and identify the affected element as pending or stale.

Cohort engagement recognition

• •Present an opt-in, batch-scoped ranking of learners by recorded completion metrics such as module

completion and attendance; participation is off by default and never affects any mandatory requirement, assessment, or credential.

• •Scope the ranking to the learner's own batch only; no cross-institution, cross-cluster, or national ranking

is presented anywhere on the platform.

• •Display a learner who is a minor by first name and initial, or by a self-selected pseudonym, never by full

name or any identifying detail beyond what the learner's own batch-mates already see in the classroom; enrollment in this feature for a minor follows the same age-appropriate consent pathway already specified in Pillar 7.

• •This capability computes a ranked display from recorded data only; it performs no inference, prediction,

or behavioral analysis, and is accordingly not subject to the AI Governance classification in Section 6.8 or listed in Annexure B.

Acceptance CriteriaPDF p.44

• •The learner home loads the latest successfully reconciled operational state and identifies any pending

external dependency.

• •The learner sees only the learner's own record; trainer and institution scope is limited to assigned

learners and the institution.

• •Lifecycle status agrees with the underlying progress, assessment, and credential records.
• •A learner enrolled in more than one course can select and view each journey independently.
• •Every displayed metric can be traced to the operational record from which it was derived.

•

Validation RulesPDF p.44

• •No dashboard field may become a separate source of truth; every status is derived from an operational

record or an explicitly stored workflow state.

• •Predictive or adaptive recommendations are visually distinguished from mandatory next actions and

recorded outcomes.

• •Program-wide aggregation is not performed in this component.
• •A view shall not return data outside the authenticated user's entity scope.
• •A stale externally sourced value shall be labeled with the time of its last successful synchronization.

Exception ManagementPDF p.44

BPMN / process flowPDF p.45

Tap to enlarge

4.9 Pillar 9: Interoperability and IntegrationPDF p.45

This component provides the integration gateway, the canonical identifier model, the learning record store, and the reconciliation controls that make both a connected-platform partner and a content-only partner safe to admit. It also governs integration with the approved identity channel, the designated awarding and examination services, credential services, the National Career Service, Skill India Digital Hub, Bhashini/AI4Bharat language services, and opportunity services. The requirements are expressed against open standards and published interfaces so that any conformant platform, whether open-source, commercial, or partner-built, can be admitted through the same route.

Functional CapabilitiesPDF p.45

Integration gateway

• •Authenticate the external system, confirm that the

requested interface and dataset fall within the approved scope, and map identifiers to the canonical model.

• •Validate each payload against the agreed

contract, apply duplicate and ownership checks, manage retries and quarantine, and stamp source and freshness on every accepted record.

The table below sets out the required integrations, their purpose, and the information exchanged.

Integration Purpose Data / Information Exchanged

Partner onboarding

• •Register a connected-platform partner's interface, scope, and data contract, and configure the

connector, credentials, and identifier mapping.

• •Register a content-only partner's submission together with its rights and metadata and register the

package or launch method.

• •Require academic review and approval of the course and content mapping before activating the

integration or publishing the content.

Acceptance CriteriaPDF p.47

• •A record imported from a connected platform reconciles to the same learner, course version, and batch

without duplicate entry.

• •Conformance of each supported standard - LTI 1.3, SCORM, xAPI, cmi5, QTI 3.0, OneRoster, and Open

Badges - is demonstrated against a reference implementation before a partner connector is accepted.

• •A partner operating its own learning platform and a partner contributing content alone can each be

onboarded end to end in a pre-production environment.

Validation RulesPDF p.47

• •An unmapped or ambiguous partner identifier is held as an exception rather than published.
• •A payload containing fields outside the agreed data contract is rejected, not partially accepted.
• •The external AI model provider integration is abstracted behind the internal AI Gateway; no pillar calls

an external model provider directly.

Exception ManagementPDF p.47

4.10 Pillar 10: AI Learning AssistancePDF p.48

The system shall provide conversational and generative assistance to learners and trainers, grounded exclusively in the approved content of the active course version, to reduce dependency on trainer availability for routine doubt resolution and to support study, revision, and content preparation. This pillar did not exist in the prior baseline; it is introduced because no other pillar in this report owns grounded conversational or generative assistance, and because the market evidence reviewed for this program (Section 4.10.5) demonstrates that such assistance is now a mature, evidenced capability in both global and Indian learning platforms.

Every capability in this pillar operates under the Advisory or Human-approved classification defined in Section 6.8. The AI Learning Assistance capability never marks an activity complete, never records an assessment outcome, and never issues or influences a credential.

4.10.1 Purpose and ScopePDF p.48

• •Purpose: to close the gap identified in Section 2.3 between a dispersed, multilingual ITI learner

population and the availability of a trainer, by providing course-grounded assistance that a learner or trainer can consult on demand.

• •Primary users: learners, for course Q&A, explanations, and study aids; trainers, for content Q&A and

session-preparation drafting.

• •Secondary users: institution administrators, who configure whether the capability is enabled for a given

course or batch.

• •Expected outcome: reduced unresolved-doubt turnaround and improved self-directed study support,

without any change to the mandatory course sequence, assessment outcome, or completion condition.

BPMN / process flowPDF p.49

Tap to enlarge

4.10.2 Functional CapabilitiesPDF p.49

AI Course Q&A Assistant and Conversational Tutor

• •Answer a learner's question using content retrieved

exclusively from the approved, published content of the learner's active course version.

• •Present the source passage or a citation to the source

alongside every generated answer.

• •Where no sufficiently relevant approved content is

retrieved, or retrieval confidence falls below the configured threshold, respond with an explicit statement that the assistant cannot answer from approved course content and direct the learner to the trainer, rather than generating an unsupported answer.

• •Support a guided, Socratic-style interaction mode for

tutoring contexts, in which the assistant offers hints, questions, and partial explanations rather than a complete solution, configurable by the institution.

• •Adapt the assistance offered to the learner's current

module and progress state within the active course.

Study and revision support

• •Generate concise summaries of approved course content

for learner revision, on request.

• •Generate flashcards from approved course content to

support spaced-repetition study, on request.

• •Generate a tailored explanation of a specific concept,

matched to the learner's demonstrated level within the course, on request.

• •Generate practice questions matched to a learner's

identified gaps, drawing on the same generation Send configurable engagement prompts, triggered by a defined condition such as elapsed inactivity since last access or an approaching assessment deadline, to encourage consistent study habits.

• •Cap prompt frequency per learner per configured period,

so encouragement does not become intrusive; a learner may disable engagement prompts entirely without affecting access to any mandatory course function.

• •Prompts draw only from the learner's own recorded

progress and course schedule; no prompt is generated from, or contributes to, a behavioral profile of a learner who is a minor, consistent with Section 6.8.

• •capability specified for formative assessment support in

Pillar 3, on request.

Trainer-facing assistance

• •Draft a session plan or supporting training material from approved course content, for trainer review

before use.

• •Answer a trainer's question about course content, sequencing, or delivery guidance, grounded in the

same approved content corpus.

BPMN / process flowPDF p.50

Tap to enlarge

• •Provide role-scoped, natural-language

access to the trainer's own assigned learner and batch data, strictly within the trainer's existing Pillar 5 and Pillar 8 access scope; this capability never exposes data outside that scope regardless of how the query is phrased.

Search and discovery

• •Provide natural-language search across

the approved course and qualification catalogue maintained in Pillar 1.

• •Provide conversational, in-course search

(Ask My Course”) that answers from the specific course's approved content, as a distinct corpus from the catalogue-wide search above.

On-demand generation

• •Generate a quiz or exercise from

approved course content on learner or trainer request, for formative, self-check use.

• •Route a request to convert an on-demand

quiz into a graded, LMS-controlled assessment through the AI-Assisted Assessment Support authoring workflow specified in Pillar 3, which requires author or assessor approval before any such item is used for a graded outcome.

4.10.3 AI Behavior, Data, and Human ControlPDF p.50

Acceptance CriteriaPDF p.51

• •Every displayed AI-generated answer is traceable to a cited source passage within the approved content

of the active course version.

• •The assistant issues an explicit unavailability response, rather than an ungrounded answer, when no

sufficiently relevant approved content is retrieved.

• •The institution can disable the AI Learning Assistance capability at course or batch level, and the

standard learner and trainer experience is unaffected when it is disabled.

• •A trainer-facing natural-language data query never returns data outside the trainer's existing assigned

scope.

• •All AI-generated content is visibly labeled as AI-generated wherever it is displayed to a learner or trainer.

Validation RulesPDF p.51

• •Content is retrieved and cited only from published content of the active course version; unpublished or

superseded content is never surfaced.

• •A response below the configured retrieval-confidence threshold is suppressed and replaced with the

unavailability response.

• •No AI Learning Assistance output can mark a module or activity complete, record an assessment

outcome, or affect a credential.

• •A quiz or exercise generated under this pillar is treated as formative unless and until it is separately

approved through the Pillar 3 AI-assisted authoring workflow.

Exception ManagementPDF p.51

Exception System Action

4.10.5 Market Evidence ConsideredPDF p.52

The introduction of this pillar is supported by evidence gathered during program research rather than by assumption. Khan Academy's Khanmigo assistant, built on a large language model and grounded in Khan Academy's own content library, uses a guided Socratic interaction design and reported substantial adoption growth across partner school districts between the 2023-24 and 2024 - 25 school years, per Khan Academy's own published account. In the Indian market, Physics Wallah's AI Guru assistant, built on a large language model with retrieval grounded in curated content, is reported by the provider to have reached a large user base within two months of launch, with a self -reported academic-query satisfaction rate; this figure is a provider claim and is presented in this report as such, not as an independently verified statistic. Skill India Digital Hub has itself introduced an AI-based conversational assistant and course/career recommendation capability at the national convergence layer, evidencing that AI-assisted learner guidance is an accepted pattern within the Government of India skilling ecosystem. These sources establish that retrieval-grounded conversational assistance is a proven, not experimental, capability class; they do not establish a specific performance guarantee for PM-SETU, which should be validated during solution design and testing (Section 7).

4.11 Pillar 11: Skills and Career IntelligencePDF p.52

The system shall maintain an evidence-based learner skills profile, anchored to National Council for Vocational Education and Training (NCVET) and National Skills Qualification Framework (NSQF) competencies, and shall provide advisory career-path, job-matching, and labor-market-informed guidance by reference to that profile and to external labor-market and opportunity data federated from the National Career Service and Skill India Digital Hub. This pillar did not exist in the prior baseline; it is introduced because course-level competency mapping in Course Management (Pillar 1) had no route to a durable, cross-course skills record or to employability guidance, despite this being central to PM-SETU's skilling-to-employment mandate.

This pillar extends, rather than duplicates, national infrastructure. Where the National Career Service or Skill India Digital Hub already performs a function - job matching, labor-market data, or career guidance at national scale - this pillar integrates with that service rather than rebuilding it, consuming and presenting federated data through the learner's PM-SETU record.

4.11.1 Purpose and ScopePDF p.52

• •Purpose: to convert recorded, in-course competency evidence into a

BPMN / process flowPDF p.53

Tap to enlarge

durable skills record that can inform a learner's next course, career path, or job match, closing the gap identified in Section 2.3 between skilling and employability.

• •Primary users: learners, for skills-profile and career guidance; institutions

and SPVs, for cluster-level skills visibility.

• •Secondary users: employers and AIPs consuming skills and job-

readiness data through the federated NCS/SIDH interface, subject to the learner's consent.

• •Expected outcome: an advisory, evidence-traceable skills and career

record that supports, but never determines, a learner's eligibility for any program component, opportunity, or credential.

4.11.2 Functional CapabilitiesPDF p.53

Skills profiling and mastery estimation

• •Build a structured skills profile for each learner from recorded competency

evidence - assessment outcomes, practical evaluations, and completed competencies - anchored to NCVET/NSQF competency definitions.

• •Estimate the learner's level of mastery for a given skill or competency from

the same recorded evidence, presented as an estimate for learner and trainer review, not a certified rating.

• •Maintain a skills and competency graph mapping skills to competencies,

courses, and qualifications, anchored to NCVET/NSQF taxonomy rather than a generic commercial taxonomy.

• •Build the profile exclusively from recorded evidence; do not infer a skill or

mastery level from unrelated personal data.

Career and employability guidance

• •Identify gaps between a learner's skills profile and a target job role’s

requirement, sourced from federated National Career Service data.

• •Recommend advisory career paths based on the learner's skills profile,

recorded interests, and learning outcomes.

• •Match completed courses and skills to relevant job and apprenticeship

opportunities, federated from the National Career Service and Skill India Digital Hub.

• •Generate an advisory job-readiness score for a defined target job role from

the learner's skills profile; this score shall never gate eligibility for any course, assessment, certification, or program component.

• •Recommend upskilling options informed by current labor-market demand

data sourced from the federate integration.

• •Provide a conversational career assistant that answers learner questions

about career and job options, grounded in the learner's own skills profile and the federated labor-market data, under the same grounding and unavailability-fallback requirements specified in Pillar 10.

Recognition-of-prior-learning support (future phase)

• •As a future-phase capability, analyze evidence submitted in support of a recognition-of-prior-learning

claim to assist, not replace, the human assessor responsible for the RPL decision under the existing SIDH RPL process.

Architecture / overviewPDF p.54

Tap to enlarge

4.11.3 AI Behavior, Data, and Human ControlPDF p.54

Acceptance CriteriaPDF p.55

• •Every skill or mastery estimate is traceable to specific recorded competency evidence.
• •Every career path or job-matching recommendation is traceable to the skills profile and, where

applicable, the federated labor-market data used to generate it.

• •A job-readiness score does not affect eligibility for any course, assessment, certification, or program

component in any pillar of this report.

• •Where the National Career Service or Skill India Digital Hub interface is unavailable, the learner's skills

profile remains visible using the last confirmed federated data, labeled as stale.

Validation RulesPDF p.55

• •The skills profile is built exclusively from recorded competency evidence within PM-SETU; it does not

combine unrelated personal data.

• •Career, job-matching, and labor-market recommendations are sourced from an explicit federated data

contract with the National Career Service or Skill India Digital Hub, never fabricated or estimated in the absence of that data.

• •A job-readiness score is presented with its underlying evidence and is never presented as a certified or

authoritative employability determination.

Exception ManagementPDF p.55

5. Solution ArchitecturePDF p.56

The PM-SETU Learning Management System shall act as a unified digital learning platform maintaining one operational learner record across a heterogeneous delivery landscape. The architecture shall enable delivery of the approved course version to an identified learner, immediate capture of learning and assessment events, governed exchange with external learning platforms and content sources, governed use of artificial intelligence within a bounded advisory scope, and role-scoped presentation of the resulting record. The architecture is organized into three layers, connected by an integration boundary, together with a cross-cutting set of AI services that every pillar draws upon rather than rebuilds:

• ●Data Layer - the sources from which learner, course, learning, assessment, skills, and credential

information originate, and the operational store that retains it, including AI interaction and provenance records.

• ●Platform Layer - the eleven functional pillars specified in Section 4, together with the cross-cutting

services, including the AI Services block, that operate across them.

• ●User Access Layer - role-scoped workspaces for learners, trainers, institutions, clusters, and

partners.

Architecture / overviewPDF p.56

Tap to enlarge

5.1 Data LayerPDF p.56

The Data Layer serves as the foundation of the learning platform. It shall collect, validate, reconcile, and standardize data from multiple internal and external sources, and shall retain the operational records required for the learning journey. This layer is critical because PM-SETU learning events originate in several places at once: natively in the platform, in institution systems, in partner-operated learning platforms, in national services that own identity, results, and credentials, and, for AI-enabled capabilities, in a governed, indexed corpus of the platform's own approved content.

Sources of inflow. Data reaches the platform from seven principal source categories (FIXED - was "six"; the table below has always had seven rows). The approved identity and enrollment channel supplies learner identity, profile, and enrollment context. The designated curriculum authority supplies authoritative trade and qualification references, and the designated examination and awarding service separately supplies formal examination, certification, and result references. Institutions and trainers enter the operational learning records generated during delivery. Industry partners either enter assigned operational records in the platform

or exchange them from a connected learning system. Content, credential, and opportunity services supply approved content metadata, credential references, labor-market data, and apprenticeship or employment hand-off information where integration has been enabled. A seventh category (FIXED - was "A sixth category"), approved AI knowledge sources, comprises the vectorized, indexed corpus of the platform's own published, approved course content, maintained separately from the raw content repository because it requires its own currency and versioning discipline distinct from the live course version.

Curriculum and qualification authority

CSTARI / DGT / NIMI / NCVET as designated

What is stored: The operational store holds the learner-to-course and batch relationship, course, content, and assessment version references, learning activity, progress, attendance, practical evidence, interventions, adaptive-learning state, skills and competency-graph state, predictive-indicator history, credential status, AI interaction and generated-content provenance records, and the source identifiers required to reconcile external records. It preserves audit history for controlled corrections and version changes. Authoritative identity source data, formal examination and certification records, and external credential documents remain in their designated systems; the platform stores only the identifiers, status, and references needed to render the learner journey and prove the source of the information.

Real-time behavior and refresh: Learner interactions generated directly in the platform - module completion, assessment attempt, trainer-recorded practical evidence, or an AI Learning Assistance interaction - shall update the operational learner record immediately after validation. Connected partner and national systems may use event-driven or scheduled synchronization according to the agreed interface. Every externally derived status shall carry the time or version of the last successful synchronization. Where a source is

unavailable, the platform shall retain the last confirmed state and identify the affected record as pending or stale rather than inferring a new outcome.

5.2 Platform LayerPDF p.58

The Platform Layer is the core application layer in which business rules, workflows, validations, learning delivery, assessment, adaptive and AI-assisted support, dashboards, and audit controls operate. Its functional components are the eleven pillars specified in Section 4. Three capabilities operate across all components: access control, which scopes every service call to the authenticated user and entity; the integration gateway, which manages external identifiers, data contracts, validation, retries, and source lineage; and the AI Services block, which provides the shared AI infrastructure every AI-enabled pillar consumes rather than rebuilds.

5.2.1 AI ServicesPDF p.59

The AI Services block provides the shared infrastructure that every AI-enabled pillar consumes. It is specified once, here, rather than rebuilt within each pillar, so that governance, versioning, and monitoring are applied consistently.

Predictive-Analytics Service

Service Function Consuming Pillars Status

Agent Orchestration / Approval Framework

No pillar calls an external AI model provider directly; every call is routed through the AI Gateway, which abstracts the underlying model provider so that no functional requirement in this report is dependent on a single named commercial AI provider, consistent with the procurement-neutrality principle applied throughout this report.

5.3 Integration and Interoperability LayerPDF p.60

The integration boundary is specified as an architectural layer rather than a set of point interfaces, because the program's delivery model depends on admitting systems that PM-SETU does not control. The gateway shall authenticate the external system, confirm that the requested interface and dataset fall within the approved scope, map identifiers to the canonical model, validate each payload against the agreed contract, apply duplicate and ownership checks, manage retries and quarantine, and stamp source and freshness on every accepted record. The required integrations, their purpose, and the information exchanged are set out in Section 4.9.

5.4 User Access LayerPDF p.60

The User Access Layer shall resolve the authenticated user to a role and entity and then constrain both navigation and data. A learner is scoped to the learner's own record. A trainer or institution administrator is scoped to the assigned institution, courses, batches, and learners. An SPV or Anchor Industry Partner is scoped to its cluster. An industry or content partner is scoped only to the course, content, on-the-job training, or integration records assigned to that partner. Centre and State users may hold governance and configuration access where formally provisioned, but program-wide monitoring is consumed through the Analytics Portal rather than through a duplicate dashboard in this platform. Access to every AI-enabled capability is scoped by the same role and entity boundaries defined here; no AI capability in this report expands a user's underlying data-access scope.

Stakeholder Workspace and Access Data Scope

Content / knowledge partner

Partner-owned resources and approved mappings

5.5 Technology PrinciplesPDF p.61

The platform should be built to current architectural practice so that it can absorb growth in learners, institutions, and integrations, and so that new delivery partners and AI capabilities can be admitted without redesign. The following principles govern solution design; they describe required properties rather than a prescribed product or vendor stack.

6. Non-Functional RequirementsPDF p.63

The PM-SETU Learning Management System will be used directly by learners, trainers, institutions, and connected partners, at scale and frequently under constrained connectivity. Its non-functional requirements therefore address responsive learner interaction, reliable capture of learning state, security of personal and academic records, availability during training and assessment periods, accessibility across device and connectivity conditions, complete auditability of changes to course, assessment, progress, and credential state, and - introduced in this version - a dedicated governance framework for every artificial intelligence capability in this report.

6.1 Performance requirementsPDF p.63

The platform shall provide a responsive and reliable experience across all stakeholder groups and shall support concurrent access without degradation. The table below sets out the minimum performance parameters under normal and peak operating conditions.

6.2 Security requirementsPDF p.65

The platform shall implement a comprehensive security framework to protect learner personal data, academic records, assessment content, AI interaction data, and partner integration data. Authentication shall be integrated with the approved government identity and access management or single sign-on mechanism. Role-based access control shall be enforced across all components, including every AI capability, so that users access only the information and functionality relevant to their designated responsibilities, and the platform shall maintain strict segregation between learners, institutions, clusters, and partners. The table below describes the security standards and frameworks required for the platform.

6.3 Availability and disaster recovery requirementsPDF p.66

The platform should be designed for high availability during normal training operations and scheduled assessment periods. A disaster recovery environment shall be maintained at a geographically separate

location to ensure continuity of critical learning operations in the event of infrastructure failure, cyber incident, natural disaster, or another disruptive event. The disaster recovery environment shall maintain synchronized copies of application data, configuration, master data, learning assets, assessment records, AI interaction and audit logs, and audit history.

Recovery procedures shall be documented, periodically tested, and governed through defined operational protocols. Recovery priorities shall place learner access, assessment-state integrity, progress data, and identity resolution ahead of non-critical administrative and AI-assisted views. Learning records committed before a service failure shall not be lost or silently duplicated on recovery. External content, credential, or AI Services integrations may be temporarily unavailable without preventing access to unaffected functionality; where an external source is unavailable, the platform shall continue to serve the last confirmed status with an indication of staleness while preventing actions that require fresh confirmation from completing incorrectly.

6.4 Backup and data retention requirementsPDF p.67

The platform shall implement automated backup mechanisms covering application databases, learning assets, assessment content and submissions, practical evidence, credential records, AI interaction and generated-content logs, master data, configuration settings, and system logs. Backups shall be performed at scheduled intervals and stored securely in accordance with government data protection guidelines. The solution shall maintain multiple backup copies to support operational recovery, disaster recovery, and regulatory compliance. Backup files shall be encrypted and protected against unauthorized access or alteration, and periodic restoration testing shall be conducted to validate recoverability and data integrity.

The platform shall preserve historical learning, assessment, progress, skills, credential, AI interaction, and workflow information for the duration prescribed under applicable government record retention policies and program governance requirements, and for the period required to substantiate any credential issued. Archived data shall remain searchable and auditable throughout the retention period. Personal data, including AI interaction data, shall be retained only for as long as the stated purpose requires, in accordance with the Digital Personal Data Protection Act, 2023, and shall follow the same retention regime as other learning records rather than a separate policy.

6.5 Audit logging and monitoring requirementsPDF p.67

Given the academic significance of the records held, comprehensive audit logging shall be a mandatory platform capability. The system shall capture authentication events, course and content version publication, learner enrollment and assignment, attendance and practical evidence entry, assessment attempts and evaluations, progress corrections, adaptive and predictive recommendations and trainer overrides, credential status changes, partner synchronization, denied access attempts, configuration changes, administrative actions, and - introduced in this version - every AI generation event, every AI-suggested score and its human confirmation, every AI-flagged plagiarism or AI-content signal and its review outcome, every predictive- indicator generation, and every career or skills recommendation.

All logs shall include user or source identification, timestamp, action performed, affected learner, course, or object, previous value, updated value, model or rule version in force at the time (where an AI capability is involved), and the origin of the transaction where applicable. Audit logs should be tamper-resistant and accessible only to authorized administrators and auditors. The platform shall support centralized log monitoring, exception detection, activity tracing, and forensic investigation. Log retention periods shall align with program audit requirements and statutory governance obligations.

6.6 Accessibility and inclusion requirementsPDF p.67

Because the platform is used directly by learners rather than only by administrators, accessibility is a functional requirement rather than a compliance formality. Learner-facing interfaces, including every AI-

enabled interface, shall conform to WCAG 2.1 Level AA and to the Guidelines for Indian Government Websites 3.0. Interfaces shall support keyboard navigation, screen-reader compatibility, sufficient color contrast, resizable text, and predictable navigation.

The platform shall support regional languages for both interface and content, including transcripts, captions, and subtitling where the underlying resource permits, delivered through the Multilingual/Speech Service (Section 5.2.1) integrated with Bhashini and AI4Bharat. It shall be usable on the low-cost mobile devices commonly available to learners, and shall operate under intermittent connectivity through download, offline activity capture, and deferred synchronization. Where a learner requires an accessible alternative to a resource, the platform shall present the alternative in place of, rather than in addition to, an inaccessible one. The AI Learning Assistance capability (Pillar 10) shall support text-based interaction in the platform's supported regional languages, with voice interaction supported where the Multilingual/Speech Service confirms adequate accuracy for the requested language.

Where a learner requires an assessment accommodation on account of disability, including extended time, an alternate format, or a sign-language or scribe accommodation for a practical or viva component, the platform shall support recording and applying that accommodation to the learner's assessment configuration without altering the assessment's content or pass criteria for other learners.

6.7 Compliance requirementsPDF p.68

The platform shall adhere to the governance principles and operational controls defined under the PM-SETU program and shall support compliance with Government of India digital governance standards, approved cybersecurity guidelines, data protection requirements, applicable education and skilling regulations, and the artificial intelligence governance requirements set out in Section 6.8. The platform shall maintain the source and authority of course, result, credential, and skills/career records so that a platform workflow, human or AI- assisted, does not supersede statutory academic or certification governance.

The platform shall maintain immutable audit trails for all academic decisions and learning transactions, including every AI-assisted decision support interaction, to support internal audit, external audit, and independent verification. All workflows related to course publication, content approval, assessment, AI- assisted authoring and scoring support, progress correction, completion, and credential issuance shall be digitally recorded and retained to ensure accountability and transparency across the PM-SETU ecosystem. Where a multilateral development partner has fiduciary or reporting requirements bearing on learning outcomes, the platform shall provide evidence-based reporting sufficient to satisfy them.

A Data Protection Impact Assessment shall be conducted prior to large-scale processing of learner personal data under this platform, with particular attention to processing involving minors, consistent with the Digital Personal Data Protection Act 2023, and the Digital Personal Data Protection Rules, 2025. The assessment shall be completed and reviewed before production go-live and revisited whenever a material change to AI processing, data collection, or a new integration is introduced.

Consistent with the Digital Personal Data Protection Act, 2023, an adult learner shall be able to request access to, correction of, and erasure of their own personal data held by the platform, and to register a grievance, through a defined process distinct from the program-level data portability and exit provisions described in Section 5.5 Nomination of a legal heir or representative for a learner's data, where applicable under the Act, shall follow the same process.

6.8 Artificial Intelligence GovernancePDF p.68

This section generalizes the governance principle first established for Personalized and Adaptive Learning (Pillar 4) - advisory-only output, explainability, human override, and prospective-only versioning - into a single

framework applicable to every artificial intelligence capability specified in this report: Pillar 3 (AI-Assisted Assessment Support), Pillar 4, Pillar 5 (AI Analytics and Prediction), Pillar 10 (AI Learning Assistance), Pillar 11 (Skills and Career Intelligence), and the Admin AI Service (Pillar 8 presentation). No principle already established for Pillar 4 is weakened by this section; every requirement below either restates an existing Pillar 4 principle at broader scope or introduces a control the absence of which would leave an AI-enabled pillar ungoverned.

6.8.1 AI Output ClassificationPDF p.69

Every output produced by an AI capability in this report shall be classified into one of four tiers, and the applicable tier shall be evident to the user from the interface presentation:

• •Advisory - informs a decision but has no direct effect on any record; the default classification for

recommendations, predictions, search results, and conversational output.

• •Human-approved - becomes usable only after an explicit human confirmation action is recorded (for

example, an AI-generated question entering a live question bank, or an AI-suggested score being confirmed by an evaluator).

• •System-enforced - an automated action taken without a per-instance human decision; no AI capability

in this report is currently specified at this tier for any function affecting curriculum, assessment outcome, completion, or credential.

• •Authoritative - reserved exclusively for results and credentials sourced from a designated external

authority (the awarding body, the credential service); no AI capability in this report generates or occupies this tier.

6.8.2 Cross-Cutting Governance ControlsPDF p.69

6.8.3 Regulatory and Standards ReferencesPDF p.70

The following legal requirements, government guidance, standards, and industry practices inform the AI Governance framework. Distinctions between binding legal requirements and non-binding guidance or benchmarks are maintained throughout this report and are not to be conflated.

• •Legal requirements: the Digital Personal Data Protection Act, 2023, and the Digital Personal Data

Protection Rules, 2025, govern lawful processing, purpose limitation, data minimization, and, under Rule 10, verifiable parental consent and the prohibition on tracking, behavioral monitoring, and profiling of children.

• •Legal requirement: the Information Technology Act, 2000, and CERT-In directions govern reasonable

security practices and incident reporting, applicable to AI Services as to any other platform component.

• •Government guidance: the NITI Aayog Responsible AI principles (safety and reliability, equality,

inclusivity and non-discrimination, privacy and security, transparency, accountability, and the protection and reinforcement of positive human values) inform the design of every AI capability in this report.

• •Standard (benchmark, not mandated for this program): ISO/IEC 42001, the international standard

for AI management systems, is referenced as an aspirational maturity benchmark for the platform's AI governance operations, and as a target certification for later program phases rather than an initial go- live requirement.

• •Standard (benchmark, not binding in India): the European Union's Artificial Intelligence Act

classification of educational and assessment AI as high-risk is used in this report as an international rigor benchmark when specifying controls for AI-Assisted Assessment Support (Pillar 3) and AI Analytics and Prediction (Pillar 5); it is not binding law for this program and is not cited as such.

• •Industry best practice: grounding generative output in retrieval from an approved content corpus

(RAG) and requiring evaluator confirmation before an AI-suggested score is recorded, reflect practice observed across the LMS and EdTech platforms reviewed for this program and are adopted here as proposed requirements.

• •Standard (benchmark, not mandated for this program): the United States NIST AI Risk Management

Framework 1.0 is referenced as an additional international benchmark for structuring AI risk identification, measurement, and management practices across the platform's AI Services.

• •Government guidance (international): UNESCO's Guidance for Generative AI in Education and

Research informs the report's approach to grounding, transparency, and human oversight specifically for generative AI capabilities used in a learning context.

6.8.4 Excluded and Deferred AI CapabilitiesPDF p.71

Two capabilities identified during market research are explicitly excluded from this program, documented here rather than silently omitted, so that the exclusion is understood as a deliberate governance decision rather than an oversight.

A further set of capabilities - AI-assisted remote proctoring, oral or speech-based assessment scoring, recognition-of-prior-learning evidence analysis, natural-language platform navigation, and autonomous multi- step AI agents for learner support, content authoring, administration, or system reconciliation - are identified as future-phase capabilities. These are not specified for initial delivery because the available evidence at the time of this report does not yet establish sufficient maturity, governability, or legal clarity (in the case of proctoring and speech-based scoring) for a Government of India program, or because the underlying agentic technology class is still an early, actively evolving market segment as of 2026. Where a future-phase capability is pursued in a later program phase, it shall be brought within the same AI Governance framework specified in this section before it is enabled for any learner, trainer, or administrator.

Where this report or an accompanying reference document identifies a specific count of artificial intelligence capabilities considered for the program, that count is 74 distinct capabilities, evaluated and classified as Core (34), Enhanced (30), Future-phase (8), or Excluded (2), as set out in Annexure B. This is the sole authoritative count for this program; any earlier working count referenced during requirements development is superseded by this figure.

6.8.5 Basis for the 74-Capability ScopePDF p.72

Market and research inputs prepared during requirements development surveyed a broader universe of 110 candidate AI capabilities across seventeen categories before this report's 74-capability scope was fixed. The reduction from 110 to 74 is explained here, at category level, so that a reviewer comparing this report against the earlier research inventory can distinguish a considered scoping decision from an omission. No capability in the broader inventory was silently dropped: each was either retained individually within the pillar structure of this report, folded into a non-functional or governance requirement that applies more strongly than a discretionary feature would, or explicitly deferred to a future phase for the reason stated.

7. Testing Strategy and Deployment ApproachPDF p.74

The PM-SETU Learning Management System shall adopt a structured testing and deployment framework to ensure that all learning, assessment, progress, credential, AI, integration, and reporting functionality operates as intended prior to production rollout. Testing is directed at three things in particular: the integrity of the learner journey, the separation of authority between the platform and the systems connected to it, and the correct, bounded operation of every AI capability against the governance framework in Section 6.8. The solution shall be considered ready for production only when the same learner, course version, assessment, and credential state can be reconciled across the screens, operational records, and external integrations that contribute to it, when each role can see and act only within its permitted scope, and when every AI capability demonstrably operates within its governed, advisory or human-approved boundary.

7.1 Testing strategyPDF p.74

The testing lifecycle shall include Unit Testing, System Integration Testing, User Acceptance Testing, Performance Testing, Security Testing, Accessibility Testing, AI Governance Testing, and Production Validation Testing.

• •Functional and journey testing. End-to-end scenarios shall cover learner activation, course and batch

assignment, digital and blended learning delivery, practical evidence capture, assessment, AI-assisted question authoring and scoring support, adaptive and predictive recommendation, AI Learning Assistance interactions, skills and career intelligence output, progress, completion, credential display, and opportunity hand-off. Negative cases shall confirm that missing prerequisites, incomplete assessment evidence, or an absent authoritative result prevent premature completion.

• •Integration and reconciliation testing. Identity connected partner platform, examination result,

credential, National Career Service/Skill India Digital Hub, Bhashini/AI4Bharat, and opportunity interfaces shall be tested for identifier mapping, duplicate handling, retries, stale data, correction, and source lineage. A record imported from a connected platform shall reconcile to the same learner, course version, and batch without duplicate entry. Conformance of each supported standard - LTI 1.3, SCORM, xAPI, cmi5, QTI 3.0, OneRoster, and Open Badges - shall be demonstrated against reference implementation before a partner connector is accepted.

• •Access-control and privacy testing. Learner, trainer, institution, SPV, industry partner, content

partner, and privileged administrator roles, including role-based AI permissions, shall be tested against permitted data scopes. Tests shall include direct service invocation and address manipulation attempts, not only screen navigation, and shall confirm that denied access is logged.

• •Assessment integrity testing. Attempt creation, autosave, submission, scoring, practical evaluation,

correction, and reattempt rules shall be validated under normal and interrupted sessions. Concurrent submission shall not create a duplicate result, an externally authoritative result shall not be recalculated, and no AI-suggested score shall become the recorded result without a captured evaluator confirmation.

• •AI governance testing. For every AI capability: confirm the output classification (Advisory / Human-

approved) matches the specification in Section 6.8.1; confirm every output is explainable and traceable to its source signal or content passage; confirm human override is available, requires a reason, and is logged; confirm model-version recoverability and prospective-only application of a configuration change; and confirm graceful degradation on AI service failure. This category comprises five named, separately tracked test types, each with its own pass criteria recorded in the test plan:

• –Grounding-rate testing: for every generative capability in Pillar 3 and Pillar 10, sample a defined

set of responses and confirm each is traceable to a cited source passage in the approved, published content of the active course version; the grounding-rate threshold is set during solution design per Section 7.2.

• –Hallucination-audit testing: independently review a sampled set of AI-generated answers,

summaries, questions, and career or job-matching recommendations against their cited source or evidence, and confirm that the unavailability fallback, rather than an ungrounded answer, is returned whenever retrieval confidence is below the configured threshold.

• –Bias and disparity-audit testing: for every predictive indicator (Pillar 5), AI-assisted assessment

support output (Pillar 3), and skills or career recommendation (Pillar 11), test outcome distribution across language, region, gender, and institution type, and confirm that a periodic disparity review is scheduled and its findings are recorded, consistent with Section 6.8.2.

• –Human-override-logging testing: for every AI capability at the Human-approved output tier,

confirm that an override or rejection cannot be submitted without a recorded reason, and that the override, its reason, the original AI output, and the human decision are all captured in the audit log defined in Section 6.5.

• –Assessment-integrity-flag testing: for plagiarism detection and AI-generated-answer detection

(Pillar 3), confirm that a flag alone never triggers a penalty, grade change, or disciplinary workflow step without a recorded evaluator decision, given the documented elevated false-positive rate of AI- content detection against non-native English writers noted in Section 4.3.

• •Accessibility and device testing. Learner journeys, including AI Learning Assistance and multilingual

interactions, shall be tested on the supported mobile and desktop profiles, with keyboard navigation and screen-reader paths, and under low-bandwidth and offline scenarios covering content fallback, resume, and deferred synchronization.

• •Performance, security, and resilience testing. Load, peak assessment concurrency, AI Gateway

load, integration-failure, backup and restore, and failover scenarios shall be executed against the minimum requirements at Section 6. Vulnerability Assessment and Penetration Testing, including of the AI Gateway, shall be completed by a CERT-In empaneled agency before production go-live.

User Acceptance Testing shall be conducted with nominated representatives from the SPMU, State authorities, SPVs, institutions, trainers, learners, and integration partners, validating actual business scenarios including course activation, content delivery, assessment administration, AI-assisted authoring and study support, progress verification, credential presentation, and partner reconciliation. Successful completion of User Acceptance Testing shall constitute formal business sign-off for production deployment.

7.2 Acceptance criteriaPDF p.75

Production deployment approval shall be granted only after successful completion of all planned testing activities and satisfaction of the following criteria.

• •All critical learner journeys pass end to end, with course version, progress, assessment, and credential

state reconciling to the underlying operational records.

• •Role and entity access tests pass for every stakeholder role, including prevention and logging of

unauthorized learner-data access and unauthorized AI-capability access.

• •Connected partner and national-system interfaces, including the National Career Service, Skill India

Digital Hub, and Bhashini/AI4Bharat, pass identifier mapping, retry, duplicate, correction, and stale-data scenarios, and standards conformance is demonstrated for each interface in scope.

• •A partner operating its own learning platform and a partner contributing content alone have each been

onboarded end to end in a pre-production environment.

• •No critical or high-severity defect remains open in learner identity, assessment integrity, progress

calculation, credential status, data isolation, or AI governance controls.

• •Every AI capability specified for initial delivery demonstrates its specified output classification,

explainability, human-override, and graceful-degradation behavior under test.

• •Grounding-rate testing, hallucination-audit testing, bias and disparity-audit testing, human-override-

logging testing, and assessment-integrity-flag testing, as defined in Section 7.1, have each been executed and their results recorded in the test plan; no AI capability specified for initial delivery is exempted from this set.

• •Performance, accessibility, and security requirements at Section 6 are met under the agreed test load

and supported device profiles.

• •All application security findings categorized as critical or high risk, including any AI Gateway findings,

are closed and retested.

• •User Acceptance Testing sign-off is obtained from nominated learners, trainers, institutions, and

integration representatives.

• •Specific numeric acceptance thresholds for AI grounding rate, hallucination rate, prediction-disparity

tolerance, and AI response-time targets not already fixed in Section 6.1 are to be defined during solution design, prior to User Acceptance Testing, and recorded in the test plan.

7.3 CI/CD and deployment approachPDF p.76

The solution shall adopt a continuous integration and continuous delivery approach to support controlled, repeatable, and auditable software releases. Source code repositories, automated build pipelines, quality checks, automated test suites, security scanning, and deployment workflows shall be integrated into a centralized DevSecOps framework. Any change to an AI model or rule configuration shall pass through the same pipeline gates as an application code change, consistent with the Model Registry's prospective-only versioning requirement (Section 6.8.2).

The deployment lifecycle shall follow a phased environment promotion model comprising Development, Testing, User Acceptance Testing, Pre-Production, and Production environments, together with a partner sandbox in which an integration partner can certify a connector before production activation. All releases shall pass through automated code validation, build verification, security scanning, and testing gates before promotion. Course and content master data, learner identity mappings, the AI knowledge-based index, and integration configuration required for the initial rollout shall be validated before production activation, and interfaces to external systems, including AI model providers, shall be enabled by environment and by partner so that a failed integration does not block independent testing.

Production rollout shall be phased according to the agreed PM-SETU implementation plan, beginning with the initial Hub-and-Spoke clusters and expanding after reconciliation, user support, and performance are proven. AI capabilities within the Core and Enhanced classification (Annexure B) shall be enabled progressively per pillar, following the same phased rollout discipline as the underlying pillar, rather than being enabled platform-wide on day one. Each release shall support controlled rollback of application code, configuration, and AI model/rule versions without rolling back valid learner transactions already committed. Post-deployment validation shall confirm login, course access, content delivery, learning -event capture, assessment state, AI capability availability and governance behavior, external synchronization, and audit logging before the release is declared live. This approach will ensure minimal deployment risk, faster release cycles, improved software quality, and complete traceability of changes throughout the lifecycle of the platform.

7.4 Product Development MethodologyPDF p.76

This platform shall be developed using an iterative, sprint-based methodology, with each pillar's Base-tier capabilities delivered and validated before Middle-tier work begins on that same pillar, consistent with the tiered rollout structure in the Comprehensive Feature Register and the advancement gating in Annexure F. A product roadmap shall be maintained and reviewed by the SPMU (Annexure E) at a cadence of [client to

confirm], with each review assessing tier-advancement readiness against the completion and reliability thresholds specified in Annexure F, rather than against elapsed time alone.

Production rollout shall be phased according to the agreed PM-SETU implementation plan, beginning with the initial Hub-and-Spoke clusters and expanding after reconciliation, user support, and performance are proven. AI capabilities within the Core and Enhanced classification (Annexure B) shall be enabled progressively per pillar, following the same phased rollout discipline as the underlying pillar, rather than being enabled platform-wide on day one. Each release shall support controlled rollback of application code, configuration, and AI model/rule versions without rolling back valid learner transactions already committed. Post-deployment validation shall confirm login, course access, content delivery, learning -event capture, assessment state, AI capability availability and governance behavior, external synchronization, and audit logging before the release is declared live. This approach will ensure minimal deployment risk, faster release cycles, improved software quality, and complete traceability of changes throughout the lifecycle of the platform.

8. Process Designs and MockupsPDF p.77

This section presents the core processes executed by the PM-SETU Learning Management System, focused on the operational hand-offs that the platform must control identity and enrollment, course and content delivery, assessment and adaptive/AI-assisted learning, progress and credentials, partner integration, AI learning assistance, predictive intervention, and skills and career intelligence. The user journeys are described at Section 3 and are not repeated here. Each subsection below presents a BPMN 2.0 process design followed immediately by its corresponding wireframe mockup, so that the process and the resulting screen are read together; the consolidated interface requirements that apply across all of them are set out in Section 8.9. The written process description, the architecture in Section 5, and the functional requirements in Section 4 describe the same system as the diagrams; no diagram introduces a capability not already specified in those sections.

8.1 Learner access, enrollment, and course activation process designPDF p.77

• 1.The learner is resolved against the approved identity and enrollment channel (Pillar 7); the platform does

not create a new identity.

• 2.The institution activates the approved course version and creates or assigns the learner to a batch (Pillar

1).

• 3.The learner's home workspace is provisioned, showing the active course, batch, and institution (Pillar

8).

• 4.Where the AI Learning Assistance capability is enabled for the course, the assistant entry point is

provisioned as part of the same workspace, with its availability state (enabled/disabled) drawn from course configuration.

BPMN / process flowPDF p.78

Tap to enlarge

The mockup below shows the resulting learner home workspace referenced in the process design above, including the next required action, the advisory recommendation card, and the AI Learning Assistance entry point.

Wireframe / screenPDF p.78

Tap to enlarge

8.2 Course and content delivery process designPDF p.79

• 5.The trainer or the platform presents the approved content sequence for the active course version (Pillar

2).

• 6.The learner consumes instructor-led, self-paced, or practical content; activity is captured against learner,

course version, and batch.

• 7.Where the learner requests it, the AI Course Q&A Assistant answers from the approved content of the

active course, citing its source, or returns an explicit unavailability response (Pillar 10); this interaction is logged but does not itself update module-completion state.

• 8.Module completion is recorded once the configured completion condition for that module is satisfied,

updating the Pillar 5 progress record.

BPMN / process flowPDF p.79

Tap to enlarge

The mockup below shows the resulting course content player referenced in the process design above, including the progress module strip, the completion control, and the advisory study-aid panel grounded in the same course content.

Wireframe / screenPDF p.80

Tap to enlarge

8.3 Assessment and AI-assisted adaptive learning process designPDF p.80

This process links assessment with the next permitted learning action and is extended in this version to show where AI-assisted authoring and evaluator support enter the flow, without altering the existing separation between the recorded academic outcome and the advisory recommendation.

Where the course is configured to require it, a baseline assessment is presented and recorded at course activation, before the first module is delivered; the same assessment instrument, or a matched equivalent, is presented again as the endline assessment once completion criteria are otherwise satisfied.

• 9.Where enabled, an author or assessor drafts a question or test using AI-assisted generation; the item is

reviewed and approved before entering the live question bank (Pillar 3).

• 10.The platform presents the assessment or practical rubric defined for the learner's course version and records the learner attempt or the authorized practical evaluation.
• 11.Where the assessment is formally owned by an external examination system, the platform consumes the result received from that source rather than recalculating it; otherwise, it scores the attempt against the configured rule.
• 12.For a descriptive answer, an AI-suggested score and rationale may be displayed alongside the evaluator's own scoring control; the evaluator's confirmed score, and only that score, becomes the recorded result.
• 13.If the configured threshold is met, the result updates the learner progress record (Pillar 5).
• 14.If it is not met, the platform may recommend remedial content, practice, or a personalized study plan (Pillar 4) and create an intervention item for the trainer, who may accept, defer, or replace the recommendation.
• 15.After the learner completes the remedial activity, reassessment proceeds under the configured attempt rule.

The AI-assisted output at every step above is advisory or human-approved and is visually and structurally separate from the recorded academic outcome and the practical evaluation panel; the reattempt eligibility table is governed by the configured attempt rule rather than by completion of the recommended remedial activity, preserving the process logic of the prior baseline.

BPMN / process flowPDF p.81

Tap to enlarge

The mockup below shows the resulting assessment screen referenced in the process design above, with the advisory recommendation panel presented separately from the mandatory question and the recorded result.

Wireframe / screenPDF p.81

Tap to enlarge

8.4 Learner progress, certification, and digital credential process designPDF p.82

• 16.Validated learning, attendance, practical, and assessment events update the consolidated Pillar 5 progress record.
• 17.The Pillar 5 AI Analytics and Prediction capability computes advisory risk and completion indicators on the configured schedule, explainable by reference to the signals used.
• 18.Where an indicator flags a learner, the trainer reviews the flag and, at the trainer's discretion, records an intervention (Pillar 5); the flag itself never changes progress state.
• 19.Once completion criteria are satisfied, the platform evaluates certificate/badge eligibility and issues or maps the credential once the issuing authority confirms issue (Pillar 6).
• 20.The learner's skills profile is updated from the same recorded evidence, and any career or job- matching guidance is refreshed (Pillar 11).

BPMN / process flowPDF p.82

Tap to enlarge

The mockup below shows the resulting consolidated view referenced in the process design above, presenting progress, issued and pending credentials, and the advisory skills profile in one workspace.

Wireframe / screenPDF p.83

Tap to enlarge

8.5 Vendor LMS and content partner integration process designPDF p.83

This process covers the two principal forms of external contribution to the platform. The integration gateway first determines whether the partner operates its own learning management system. A connected-platform partner registers the interface, scope, and data contract, and the gateway configures the connector, credentials, and identifier mapping. A content-only partner submits the resource together with its rights and metadata, and the gateway registers the package or launch method. Both paths then converge on reconciliation to the PM-SETU canonical model: learner, course, version, batch, and competency identifiers are mapped while the partner's own identifiers are preserved. The academic reviewer validates the course and content mapping and, if it is not acceptable, returns it for correction without activating the integration or publishing the content. Once approved, and only once approved, published content becomes eligible for indexing into the AI knowledge base described in Section 4.10 and Section 5.1. The platform then activates the connector or publishes the content version and begins exchanging the completion, assessment, credential, or other approved events defined in the contract, with reconciliation and duplicate checks applied on every exchange.

BPMN / process flowPDF p.84

Tap to enlarge

The mockup below shows the resulting integration console referenced in the process design above, listing connector status, last synchronization, and outstanding reconciliation exceptions by partner.

Wireframe / screenPDF p.84

Tap to enlarge

8.6 AI Learning Assistance interaction process designPDF p.85

• 21.The learner or trainer submits a query through the AI Learning Assistance entry point on their existing workspace (Pillar 8, Pillar 10).
• 22.The AI Gateway routes the query to the Retrieval-Augmented Generation service, which retrieves candidate passages from the indexed, approved content of the active course version only.
• 23.If retrieval confidence meets the configured threshold, a grounded answer is generated and returned with its source citation; if not, the explicit unavailability response is returned instead.
• 24.The interaction, including the retrieved source and the response, is logged by AI Observability for audit and quality review.
• 25.The user may flag the response; a flagged response is routed to institution or program review without interrupting the capability for other users.

BPMN / process flowPDF p.85

Tap to enlarge

The mockup below shows the resulting conversational panel referenced in the process design above, including the source citation, the multilingual translation note, and the input controls.

Wireframe / screenPDF p.86

Tap to enlarge

8.7 AI predictive indicator and intervention process designPDF p.86

• 26.Validated attendance, progress, engagement, and assessment signals accumulate in the learner's Pillar 5 record.
• 27.The Predictive-Analytics Service computes advisory risk and completion indicators on the configured schedule, provided sufficient signal history exists; otherwise, the prediction is suppressed.
• 28.The indicator, with its explanation, is surfaced to the trainer and institution view (Pillar 8), visually distinguished from a recorded outcome.
• 29.The trainer reviews the indicator and, at their discretion, records an intervention, a follow-up date, and an eventual outcome (Pillar 5); no automated action results from the indicator alone.
• 30.Indicator quality and any disparity across learner groups are reviewed periodically by the model owner, consistent with Section 6.8.

BPMN / process flowPDF p.87

Tap to enlarge

The mockup below shows the resulting trainer batch view referenced in the process design above, with the advisory risk indicator column visually distinguished from recorded attendance, progress, and assessment status.

Wireframe / screenPDF p.87

Tap to enlarge

8.8 Skills and Career Intelligence process designPDF p.88

• 31.Recorded competency evidence from Pillar 3 (assessment) and Pillar 5 (progress) updates the learner's skills profile, anchored to NCVET/NSQF competency definitions (Pillar 11).
• 32.The Skills Graph Service estimates mastery for each evidenced competency and positions the learner within the skills/competency graph.
• 33.Where the National Career Service / Skill India Digital Hub interface is available, federated job, apprenticeship, and labor-market data is retrieved and matched against the learner's profile.
• 34.Advisory career path suggestions, job matches, and a job-readiness score are presented to the learner, each traceable to the specific evidence and federated data used.
• 35.The learner may query the AI Career Assistant for further guidance, grounded in the same profile and federated data, under the same grounding and unavailability-fallback rules as Pillar 10.

BPMN / process flowPDF p.88

Tap to enlarge

The mockup below shows the resulting career guidance screen referenced in the process design above, including suggested career paths, matched job opportunities federated from the National Career Service, the advisory job-readiness score, and the AI Career Assistant.

Wireframe / screenPDF p.89

Tap to enlarge

9 AnnexuresPDF p.89

The annexures below are supplementary reference material and do not introduce any requirement not already specified in Sections 1 through 8. Where a figure in an annexure appears to conflict with a figure in the main body of this report, the main body governs.

Annexure A: Persona-by-Pillar Access MatrixPDF p.89

This annexure states, for every stakeholder against every one of the eleven functional pillars, the exact access level granted. It elaborates the general access principles set out in Section 5.4 to the level of granularity required for solution design, test-plan preparation, and role-based access control configuration, and does not introduce a new access rule beyond what is specified pillar by pillar in Section 4. Where this matrix and Section 4 or Section 5.4 appear to conflict, the pillar-level requirement in Section 4 governs.

The matrix is read by row. Each cell states the access level for that stakeholder in that pillar, using the following access-level vocabulary consistently throughout: Full (create, update, and manage within the stated scope), View (read-only), Config (configuration or enablement authority without operational use), Own (limited to records the stakeholder itself owns or generated), Aggr., shown in full as Aggregate (cluster- or institution- level summary only, never an individual learner's underlying record), and N/A, shown in full as Not applicable (no access, by design, because the pillar's function does not extend to that stakeholder).

SPV/AI P

Conten t Pvt

Academic Reviewer Model Owner Integration

Full (review/appr ove only)

Pillar 2: Content Delivery Own Full Full View Own Full Config

Pillar 4: Personalized and Adaptive Learning

Pillar 6: Certification and Badges

Pillar 8: Student Lifecycle Dashboards and Management

Pillar 10: AI Learning Assistance

N/A* (Pillar 11, Industry / content partner column): Not applicable unless a specific federation agreement is in force. Where a stakeholder's access varies by scope within a pillar rather than being a single flat level, the specific scope is stated in the pillar's Functional Capabilities in Section 4 and in the narrative note below; this table states the governing access level only.

The Config access level carries a different specific scope in each pillar and is not a single undifferentiated permission: in Pillar 1 it is course-version publication and applicability; in Pillar 4 it is adaptive-recommendation model and rule configuration; in Pillar 7 it is user and identity provisioning, distinct for the ITI administrator (institution-level provisioning) and the SPV / Anchor Industry Partner (cluster-level provisioning); in Pillar 9 it is connector and mapping governance; in Pillars 10 and 11 it is AI Governance configuration under Section 6.8, including, for the ITI administrator in Pillar 10 specifically, the course- or batch-level enable and disable control described in Section 4.10.2. The specific scope of every Config, Own, View, and Aggregate entry is stated in full in the Functional Capabilities of the corresponding pillar in Section 4; this table states the governing access level for quick reference and does not restate every scope qualifier.

Two entries are highlighted here because they were the subject of specific review during requirements development. First, the SPV / Anchor Industry Partner has Not applicable access to Pillar 10 (AI Learning Assistance): this is a deliberate design choice, not an oversight, because AI Learning Assistance is a per-learner and per-trainer conversational capability with no cluster-level aggregate view defined anywhere in Section 4.10, consistent with the data-minimization principle in Section 6.8.2, which does not permit a cluster-level rollup of individual conversational interactions. Second, both the SPV / Anchor Industry Partner and the ITI administrator have Aggregate, not Own or Full, access to Pillar 11 (Skills and Career Intelligence): institution and cluster users see institution- and cluster-level skills and employability summaries presented through Pillar 8, never an individual learner's underlying skills profile or career conversation, unless the learner's own institution has separately enabled record-level sharing as stated for the trainer role.

Academic Reviewer's Full access on Pillars 2 and 9 is scoped strictly to the review/approve action on submitted content and mappings, not general content authoring or connector operation. Model Owner's Config access is scoped to AI model and rule governance specifically, not general pillar administration. These three roles are new additions to this matrix, resolving an inconsistency where each was operationally referenced in Section 4 and Section 8 but absent from this table.

Annexure B: AI Capability RegisterPDF p.92

This register resolves, as the single authoritative source, the AI capability count for this program. Seventy- four (74) distinct artificial intelligence capabilities were identified, evaluated, and classified during requirements development. This figure supersedes any differing count referenced in working or interim research material produced during development of this report. Capabilities are classified as Core (specified for initial delivery under normal governance), Enhanced (specified for initial delivery with the additional controls noted in the relevant pillar), Future (identified but deferred to a later program phase, not specified for initial delivery), or Excluded (identified and explicitly not recommended for this program, per Section 6.8.4).

9 Trainer Accept / Defer / Override Pillar 4 Core

29 Plagiarism Detection Pillar 3 Enhanced

# AI Capability DPR Location Classification

Verified: Core 34 + Enhanced 30 + Future 8 + Excluded 2 = 74, matching the stated total exactly. Sequential 1-74, no gaps or duplicates.

Annexure C: Glossary and AbbreviationsPDF p.95

AIP Anchor Industry Partner

ITI Industrial Training Institute

LMS Learning Management System

NCS National Career Service

Annexure D: Regulatory and Standards ReferencesPDF p.97

Reference Category Application in this Report

ISO/IEC 27001:2022 International standard

Annexure E: Project Governance StructurePDF p.98

This program shall be governed through a tiered structure consistent with the roles already referenced throughout this report, formalized here for clarity of reporting lines and decision authority.

Program-level governance

A State Project Management Unit (SPMU) shall hold overall program accountability, including approval of course and qualification standards changes, AI Governance framework changes per Section 6.8, and go/no- go decisions for each phased rollout described in Section 7.3. The SPMU shall be chaired by [client to confirm designated authority] and shall meet on a [client to confirm cadence, e.g. monthly] basis.

Cluster-level governance

Each SPV / Anchor Industry Partner shall be accountable for delivery coordination and reconciliation across its Hub-and-Spoke cluster, as specified in Section 3.5, reporting cluster-level status to the SPMU on a schedule to be confirmed.

Institution-level governance

Each participating ITI shall be accountable for course and batch operations, data-quality resolution, and institutional reporting, as specified in Section 4.8 and Section 4.9, with escalation to the SPV/AIP or SPMU where an exception exceeds institutional resolution authority.

AI Governance oversight

Consistent with Section 6.8, a designated AI Governance function, reporting to the SPMU, shall be accountable for model/rule version approval, periodic disparity review (Section 6.8.2), and the exclusion/inclusion decisions recorded in Section 6.8.4. This function does not sit within any single pillar and holds authority across all AI-enabled capabilities in this report.

Decision authority summary

This structure formalizes reporting relationships already implied by the roles and escalation patterns described throughout Sections 3, 4, and 8 of this report; it does not introduce a new organizational entity beyond what those sections already assume.

Annexure F: Indicative Implementation TimelinePDF p.99

The phasing principle for this program is already established in Section 7.3: rollout begins with initial Hub- and-Spoke clusters and expands after reconciliation, user support, and performance are proven, with AI capabilities enabled progressively per pillar rather than platform-wide on day one. This annexure sets out an indicative phase structure consistent with that principle. Durations shown are indicative planning assumptions, not committed dates; actual scheduling shall be confirmed by the SPMU against program funding and readiness.

Architecture / overviewPDF p.99

Tap to enlarge

Phase 2 - Core Learning Operations

Phase 4 - Enhanced Capabilities and Cluster Expansion

Enhanced-classified AI capabilities, expansion beyond initial clusters

Future-phase and excluded capabilities (Section 6.8.4, Annexure B) are intentionally not scheduled in any phase above; a Future-phase capability shall only enter a future timeline revision once it has been brought within the AI Governance framework per Section 6.8.4's own requirement.

Advancement from one rollout tier to the next is gated on demonstrated evidence, not calendar time alone. Specifically, advancement from Base to Middle tier requires a demonstrated Base-tier course-completion rate of no less than [client to confirm threshold] and an application-crash rate within [client to confirm SLA]. This gating is directly informed by external evidence: the Rajya Sabha Standing Committee on Education's report of February 2025 found that fewer than 4% of enrolees on the SWAYAM platform have completed their courses since 2017, and independent reporting on DIKSHA documents ongoing reliability complaints despite substantial reach. This report's phased advancement is accordingly evidence-gated, so that Middle- and Progressive-tier features, including AI-assisted and predictive capabilities, are not layered onto a platform whose Base-tier reliability and learner completion are not yet proven.

Annexure G: Risk RegisterPDF p.100

The risks below are drawn directly from the boundaries, fallback behaviours, and exception-management provisions already specified throughout this report, restated here as a consolidated register rather than introduced as new concerns.

Risk Source in this Report Mitigation Already Specified

Risk Source in this Report Mitigation Already Specified

Section 8.5, Section 8.6

Service unavailability during a scheduled assessment window

Section 6.1, Section 6.3

Program-level risks outside this report's technical scope (funding continuity, staffing, procurement delays)

Not addressed elsewhere in this report

This register deliberately does not include generic project-management risks (funding, staffing, procurement) that are not addressed anywhere else in the technical body of this report; adding invented mitigations for those here would misrepresent this document's actual scope. That last row is included specifically so the gap is visible, not silently absent.

Annexure H: Change Management and Capacity BuildingPDF p.101

This report's accessibility, multilingual, and human-in-the-loop requirements (Sections 6.6 and 6.8) imply a corresponding adoption and training need for each stakeholder role. This annexure states that need formally, without prescribing a training program this report has no basis to design in detail.

Trainer and evaluator readiness: Trainers and evaluators shall require orientation on the AI-assisted authoring and scoring-support workflows specified in Section 4.3 and Section 8.3, with particular emphasis on the confirm/edit/reject action required before any AI-suggested question or score is used or recorded, and on when and how to record a human override with its required reason (Section 6.8.2).

Institution administrator readiness: ITI Administrators shall require orientation on exception review (Section 8.7's advisory findings, never automatic corrections) and on the institution-level analytics introduced in Section 4.8, so that an advisory indicator is correctly understood as advisory rather than acted upon as a directive.

Learner readiness: Learner-facing AI capabilities (Pillar 10, Pillar 11) shall be introduced with in-product explanation consistent with the explainability requirement in Section 4.4, so a learner understands a recommendation's advisory nature without requiring separate training material.

Multilingual and accessibility rollout: Consistent with Section 6.6, capacity-building material itself shall be made available in the platform's supported regional languages and in an accessible format, rather than only in English.

Institution digital-readiness assessment: Given the low current digital maturity at institution level described in Section 2.2, adoption planning shall be informed by a per-institution digital-readiness assessment prior to onboarding, rather than a uniform rollout assumption across all institutions. [Client to confirm: the specific readiness index or scoring method to be used, and the threshold below which an institution receives additional onboarding support before go-live.] This assessment directly addresses the adoption gap this report's own Section 2.2 diagnoses, rather than assuming it away.

Scope note [Client to confirm ownership and detailed curriculum for this capacity-building effort; this annexure states the need this report's own requirements create, not a complete training program.]

Annexure I: Procurement ApproachPDF p.102

Consistent with the procurement-neutrality principle underlying the AI Gateway design (Section 5.2), no AI model provider, cloud provider, or system integrator is named or presumed in this report. [Client to confirm single-integrator or multi-vendor delivery model; empanelment route (e.g., GeM, MeitY panel); and the evaluation criteria by which an AI model provider is selected and rotated behind the AI Gateway.]

Any procurement decision made under this approach shall preserve the provider-abstraction principle already established in Section 5.2, so that no future capability in this report becomes dependent on a single named commercial provider.

Annexure J: Data Migration ApproachPDF p.102

Section 2.2 describes the current state at institution level as predominantly attendance registers, spreadsheets, and informal messaging-group coordination. [Client to confirm whether any existing institution- held learner history is to be migrated into PM-SETU at go-live, or whether each learner's record begins fresh at enrolment.]

Where migration is confirmed, the migrated record shall carry its original source and date, consistent with the source-lineage principle already applied to externally sourced records elsewhere in this report (Section 8.6), so a migrated record is never presented as if it were platform-generated.

Annexure K: Support and Helpdesk ModelPDF p.102

AI Support-Ticket Classification (Capability #67, Annexure B) classifies an incoming ticket; it does not resolve it. [Client to confirm the L1/L2/L3 escalation structure: L1 owner and response-time target, L2 escalation criteria and owner, and L3 (platform/technical) escalation path.]

Every ticket, regardless of tier, shall remain linked to its AI-assigned classification and any human re

classification for audit purposes, consistent with Section 6.5.

Annexure L: Comprehensive Feature RegisterPDF p.103

This register consolidates every functional capability specified in Section 4 of this report, one table per pillar, so the complete capability set can be reviewed in a single location. The "AI-Governed" column cites the exact Annexure B capability number(s) a row corresponds to. Where a single row bundles more than one individually numbered Annexure B entry, that reflects a shared capability area, not an omission - Annexure B remains authoritative for the full 74-capability count.

Methodology: Base tier capabilities are those without which the platform cannot support a real ITI batch. Middle tier capabilities meaningfully strengthen delivery once Base tier is proven working. Progressive tier capabilities are advanced, predominantly predictive or advisory AI capabilities, layered on top only once Base and Middle tier reliability is demonstrated.

Architecture / overviewPDF p.103

Tap to enlarge

Annexure B capabilities classified Future or Excluded (#31, #34, #49, #53, #62, #70-74) are intentionally not assigned a rollout tier, consistent with Section 6.8.4.

Pillar 1: Course Management

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 2: Content Delivery

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 3: Assessment and Evaluation

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 4: Personalized and Adaptive Learning

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 5: Learner Progress Tracking

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 6: Certification and Badges

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 7: Digital Identity and Credential Integration

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 8: Student Lifecycle Dashboards and Management

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 9: Interoperability and Integration

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Pillar 10: AI Learning Assistance

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Study and revision support Yes - #16, #17, #18, #19, #20 Middle

Pillar 11: Skills and Career Intelligence

Functional Capability AI-Governed (Annexure B #) Rollout Tier

#58, #59 Progressive

Platform Layer (not tied to a single pillar)

Functional Capability AI-Governed (Annexure B #) Rollout Tier

Annexure M: LMS Platform BenchmarkingPDF p.107

Part A: Basis of Comparison

This annexure benchmarks the platform described in this report against three tiers of comparator: Indian public-sector systems this platform operates alongside (Skill India Digital Hub, NIMI/Bharat Skills); the open- source technical baseline underlying comparable government deployments (Moodle); and commercial best- in-class platforms (Docebo, Talent-LMS, Canvas, D2L Brightspace, Coursera for Business). Comparator capabilities are stated as confirmed from the comparator's own published documentation as of [insert month/year at time of finalization], not inferred; where a comparator's documentation does not describe a capability, this report states "not publicly documented," not "does not support," since absence of documentation is not proof of absence.

Part B1: Government of India PlatformsPDF p.107

Capability PM-SETU SIDH NIMI DIKSHA SWAYAM Plus

Course/content management Yes Yes Repository only Yes Yes Yes

Summative assessment engine Yes, confirmed Routed to

agencies No Not core Yes No

Part B2: Open-Source and Enterprise Commercial PlatformsPDF p.107

Capability PM-SETU Moodle Docebo Canvas Brightspace

Summative assessment Yes, confirmed Yes Yes Basic Yes

QTI 3.0 Yes No, native No No No

Open Badges 3.0 Yes Plugin Roadmap Canvas Credentials

Part B3: Consumer-Facing and MOOC-Style Commercial PlatformsPDF p.107

AI Q&A grounded in content Yes Yes, Coach No No Partial

SCORM Yes N/A Not documented Via embed only Yes

xAPI Yes N/A Yes, API integration Yes, Enterprise plan Yes

Annexure N: Business Case for a Dedicated PM-SETU LMSPDF p.108

The Question

Given that Skill India Digital Hub (SIDH) already exists as a national platform, and that Bharat Skills, NIMI, DIKSHA, and SWAYAM Plus are also already operating nationally, why does PM-SETU require its own Learning Management System rather than extending SIDH or adopting one of these other existing solutions?

The Answer

Every one of these systems performs a genuine, valuable function at national scale, but none of them, individually or in combination, performs the specific function PM-SETU delivery requires: a batch-level operational teaching record, held for a specific cohort at a specific institution, with evaluator-confirmed assessment and a governed AI layer bound to that record - this is a role no existing platform occupies, not a feature gap in an otherwise-suitable one.

Part 1: Why SIDH Specifically Cannot Be Extended to Do This

This report's own Section 2.1 already establishes SIDH's role precisely, in terms worth restating here directly:

“It is not an instructional delivery system. It does not maintain the batch-level teaching record, the trainer- recorded practical evidence, the assessment attempt history, the module-level progress state, or the intervention record on which day-to-day ITI delivery depends.”

Four reasons follow from this, each a difference in system role, not a gap that a feature release would close:

• ●Operating scale and record granularity. SIDH's stated function is national discovery, identity

resolution, and career intelligence, consolidating participation data for national reporting. It operates at convergence scale - aggregating what has happened across the ecosystem - not at the granularity of a specific batch's daily attendance, a specific trainer's practical evidence, or a specific learner's module-by-module progress. Adding this granularity would not extend SIDH's role; it would require SIDH to become an operational system of record for every institution and partner nationally, which is a different platform by definition, not a new feature of the existing one.

• ●Assessment authority. SIDH has no native, evaluator-confirmed summative assessment engine;

formal assessment is routed to external NCVET agencies. PM-SETU's core requirement - LMS- controlled assessment with mandatory human confirmation of every AI-suggested score, and a governed distinction between an LMS-scored attempt and an externally authoritative result - has no equivalent anywhere in SIDH's current architecture and building one would itself constitute building a new LMS inside SIDH.

• ●AI grounding boundary. SIDH's AI capability, the Skill India Assistant (SIA), is explicitly a

recommendation and course/scheme-discovery chatbot operating at the national convergence layer. PM-SETU's requirement is an assistant grounded in the specific, active, approved content of one

learner's one course, answering from that content with a citation and an honest fallback when it cannot answer. These are different problems with different grounding data, different governance requirements, and different failure modes; they would need to coexist as separate systems even if built by the same engineering team under the same program.

• ●Access-control model. SIDH's reporting and identity functions are consumed nationally, without the

fine-grained, per-institution and per-batch access boundary this report's role-based access model requires (an ITI Administrator scoped to their own institution, an SPV/AIP scoped to their own cluster, a trainer scoped to their own batch). Retrofitting this access model onto a national convergence platform is a governance and security re-architecture, not a configuration change.

Part 2: Why the Other Existing National Platforms Also Do Not Close This Gap

The comment this annexure responds to named SIDH specifically but asked about “any other existing solution.” The three other national platforms already documented in Section 2.1 were examined on the same basis; none change the conclusion, for reasons specific to each:

Part 3: What This Costs If Left Unaddressed

Section 2.3 of this report already documents nine specifics, observed gaps in the current landscape. Each is a direct consequence of no dedicated operational system existing today, and each would persist unchanged under any of the alternatives examined above, since none of them are designed to hold an operational teaching record in the first place. Three with the most direct bearing on this decision:

• ●No system carries the learner end to end - identity sits on one platform, curriculum in another, results

in a third, and delivery in none of them. This is the direct, unavoidable consequence of relying on systems that were each built for a different, narrower purpose.

• ●The teaching record is not digitized at batch level - attendance, practical evidence, and module

completion exist nowhere as one verifiable record, because no existing national platform is built to hold one.

• ●Formal results and in-course assessment are conflated - without a system that distinguishes an LMS-

scored attempt from an externally authoritative result, the two continue to be treated as interchangeable, which this report's own evaluator-confirmation model is specifically designed to end.

Part 4: Precedent - What Fragmentation Looks Like in Practice

This report's own benchmarking work (Annexure M) surfaced a directly relevant cautionary precedent: AIIMS operates no single, unified learning platform across its campuses - Delhi, Bilaspur, Bathinda, and Rishikesh each run separate systems. This is offered here not as a comparator on capability, but as a concrete illustration of the failure mode that results when an institution or program proceeds without a single system of record: fragmentation by default, institution by institution, is what happens in the absence of a deliberate decision to build one shared operational system. PM-SETU's single-system-of-record design is a direct response to avoiding this outcome at national ITI scale.

Part 5: Why This Is Not Duplication

PM-SETU LMS federates with every platform examined above rather than replacing any part of what they do well: identity resolution defers to the approved identity channel (Section 4.7); career-path and job-matching guidance draws on SIDH and the National Career Service rather than rebuilding them (Section 4.11); formal credential custody remains with DigiLocker and the Academic Bank of Credits (Section 4.7); and curriculum and question-bank content is sourced from Bharat Skills and NIMI rather than re-authored (Section 4.1, Section 4.2). The full capability-level comparison against fifteen platforms is provided at Annexure M; this annexure states the architectural reason a separate operational system is required at all, independent of any single feature-by-feature comparison.

Part 6: What Extending SIDH Instead Would Actually Require

To make this concrete rather than asserted, retrofitting SIDH to perform PM-SETU's function would require, at minimum, all of the following to be added to a platform explicitly designed as a lightweight, federated discovery layer:

• ●Batch and cohort management, scoped per institution, for every ITI nationally - a data model SIDH

does not currently hold

• ●A native, evaluator-confirmed summative assessment engine, replacing the current routing to

external NCVET agencies

• ●Trainer-facing tools for practical-evidence capture, attendance, and intervention recording at daily

operational cadence

• ●A role-based access model scoped to institution and batch, layered onto a platform currently built for

national-scale, largely learner-facing access

• ●A course-content-grounded AI assistant, architecturally distinct from the existing national

recommendation chatbot (SIA), with its own governance and observability requirements

This is not a feature roadmap; it is a description of building a second platform inside the first one, while the first continues performing its own national function. A purpose-built LMS that federates with SIDH's existing, already-working scope is the smaller, lower-risk undertaking of the two.

Conclusion

The decision to build a dedicated PM-SETU LMS is a decision about system role, made once, at the architecture level - not a series of feature decisions that could be revisited platform by platform. Every existing national system examined in this annexure performs its own function well; none of them performs this one, and none could be made without becoming, in substance, a new system built inside an existing one.