Develop an AI enabled learning platform that identifies competency gaps, recommends personalized training through integration with the iGOT Karmayogi ecosystem, and capable of generating Quizzes and Multiple choice questions (MCQs) from uploaded learning materials to strengthen capacity building in India's Official Statistical System.

1. Executive Snapshot

This dossier provides a comprehensive technical, architectural, and strategic Blueprint for developing an AI-enabled Skill Intelligence and Learning Platform for the Ministry of Statistics and Programme Implementation (MoSPI) [PS]. The system transforms workforce capacity building across India's Official Statistical System by replacing static, rule-based training assignments with an automated, competency-driven learning engine [INFERENCE]. By analyzing employee profiles (designation, department, experience, past training) against predefined statistical and digital competency frameworks, the platform performs automated skill-gap analysis, generates personalized pathways linking iGOT Karmayogi courses with National Statistical Systems Training Academy (NSSTA) Training Programme Advisory Committee (TPAC) offerings, and features an AI-powered Intelligent Assessment Engine that auto-generates MCQs/quizzes with detailed explanations from uploaded multi-format courseware [PS]. Built as an interoperable, cloud-ready web application with Role-Based Access Control (RBAC), Single Sign-On (SSO), and real-time predictive analytics, this platform enables MoSPI’s Data Informatics & Innovation Division (DIID) to proactively address emerging skill gaps in AI/ML, Big Data, and advanced analytics [PS, INFERENCE].

Fundamental System Objective: To deliver a secure, scalable, cloud-ready AI skill intelligence platform that dynamically maps employee profiles to official statistical competency frameworks, automates skill-gap identification, generates personalized iGOT/NSSTA pathways, and generates validated assessments from multi-modal learning resources to build a future-ready statistical workforce [PS, INFERENCE].

2. Problem Deconstruction

The problem statement supplied in the SIH JSON is deconstructed into atomic, explicit requirements and derived technical implications below:

• REQ-01: Automated Employee Competency Profiling

  • Source Text: "The system should automatically create a comprehensive competency profile for every official using information such as designation, department, job role, current assignment, educational qualifications, work experience, and previous trainings." [PS]
  • Interpretation: The platform must ingest structural HR data and build a normalized schema representing each official's background [INFERENCE].
  • Implementation Implication: Requires an employee data intake pipeline (via HRMS API or CSV import) and an automated profile construction service [INFERENCE].
  • Confidence: [PS]

• REQ-02: Predefined Competency Evaluation & Skill-Gap Analysis

  • Source Text: "Based on this profile, the platform should evaluate the official's existing competencies against predefined competency frameworks for Official Statistics and identify knowledge and skill gaps." [PS]
  • Interpretation: Quantitative evaluation comparing evaluated employee skills against a standard target matrix [INFERENCE].
  • Implementation Implication: Implementation of a matrix-matching algorithm calculating delta scores ($Gap = Competency_{Target} - Competency_{Current}$) [INFERENCE].
  • Confidence: [PS]

• REQ-03: Multi-Domain Competency Taxonomy Mapping

  • Source Text: "The AI engine should map competencies across multiple domains, including: Statistical Competencies (Survey Design, Sampling, National Accounts, Price Statistics, Labour Statistics, Agricultural Statistics, Industrial Statistics, SDG Indicators, Metadata Standards, Data Quality Frameworks), Technical Competencies (Python, R, SQL, Stata, SPSS, SAS, GIS, Data Visualization, AI/ML, Cloud Computing, APIs, Open Data), Digital Governance (Cybersecurity, Data Privacy, Digital Signatures, Government Cloud, DPI), Behavioural & Managerial (Leadership, Communication, Project Management, Ethics, Decision Making, Change Management)." [PS]
  • Interpretation: The system taxonomy must strictly model these four primary pillars and their sub-domains [INFERENCE].
  • Implementation Implication: Graph/vector taxonomy store representing relationships between job roles, required proficiency levels (1–5), and sub-competencies [INFERENCE].
  • Confidence: [PS]

• REQ-04: Personalized Learning Pathways & Multi-Source Recommendation Engine

  • Source Text: "Using AI techniques such as Machine Learning, Natural Language Processing (NLP), Large Language Models (LLMs), semantic search, and competency mapping, the platform should recommend personalized learning pathways from the iGOT Karmayogi course repository... as well as NSSTA’s TPAC recommended Training Programme." [PS]
  • Interpretation: Dual-channel recommendations combining external iGOT digital content with internal/residential NSSTA TPAC calendar courses [INFERENCE].
  • Implementation Implication: Hybrid semantic search & vector recommendation pipeline scoring courses based on skill gap delta, career level, and departmental priority [INFERENCE].
  • Confidence: [PS]

• REQ-05: Seamless iGOT Karmayogi Ecosystem Integration

  • Source Text: "The platform should integrate seamlessly with iGOT Karmayogi APIs to retrieve course catalogues, recommend relevant courses, monitor enrolment and completion status, and update competency scores automatically." [PS]
  • Interpretation: Bidirectional API sync with Karmayogi Bharat/iGOT infrastructure [INFERENCE].
  • Implementation Implication: REST/GraphQL API integration layer with OAuth2 authentication, webhook listeners for course completion events, and automated profile updates [INFERENCE].
  • Confidence: [PS]

• REQ-06: AI-Powered Virtual Assistant & Adaptive Support

  • Source Text: "To support continuous learning, the solution should provide AI-powered virtual assistants for learner support, adaptive assessments, interactive learning modules, virtual laboratories, quizzes, and multilingual learning resources." [PS]
  • Interpretation: 24/7 contextual learner assistant capable of answering domain questions and dynamically adapting quiz difficulty [INFERENCE].
  • Implementation Implication: RAG (Retrieval-Augmented Generation) chatbot trained on statistical documentation and adaptive item-response theory (IRT) engine [INFERENCE].
  • Confidence: [PS]

• REQ-07: AI Intelligent Assessment Engine (Multi-Format MCQ/Quiz Generation)

  • Source Text: "AI powered Intelligent Assessment Engine capable of generating objective type questions (MCQs), and quizzes from uploaded learning materials such as documents, presentations, videos etc. It should provide instant evaluation, explanations for correct answers, and personalized feedback... leverage Large Language Models (LLMs), Natural Language Processing (NLP)." [PS]
  • Interpretation: Ingestion engine for PDF/PPTX/MP4/DOCX files that extracts content, processes transcripts, and generates structured evaluation items [INFERENCE].
  • Implementation Implication: Document parsing pipeline (Unstructured/PyMuPDF), video transcription (Whisper/VTT parsing), LLM prompt template pipelines with JSON schema enforcers, and instant evaluation logic [INFERENCE].
  • Confidence: [PS]

• REQ-08: Dual Interactive Analytics Dashboards (Learner & Administrator)

  • Source Text: "The employee dashboard should display current competency levels, identified skill gaps, recommended learning paths, learning hours, and overall progress. The administrator dashboard should provide organization-wide insights into workforce competencies, training effectiveness, competency distribution, emerging skill requirements, and predictive analytics for future capacity-building needs." [PS]
  • Interpretation: Role-differentiated analytical views with real-time tracking and predictive workforce forecasting [INFERENCE].
  • Implementation Implication: Responsive web analytics dashboards with visual gap radar charts, aggregation pipelines, and predictive trend modeling [INFERENCE].
  • Confidence: [PS]

• REQ-09: Enterprise Security, Access Control, and Interoperability

  • Source Text: "The solution should be designed as a secure, scalable, cloud-ready, and interoperable web platform... support role-based access control, Single Sign-On (SSO), and secure data exchange while ensuring compliance with government cybersecurity and data privacy guidelines." [PS]
  • Interpretation: Deployment-ready architecture compliant with CERT-In guidelines, OAuth2/SAML SSO integration (e.g., Parichay/NIC SSO), and strict RBAC [INFERENCE].
  • Implementation Implication: JWT/OAuth2 authentication, AES-256 data encryption at rest, TLS 1.3 in transit, and role middleware [INFERENCE].
  • Confidence: [PS]

3. Stakeholders & User Needs

4. Requirements Traceability

5. Constraints & Environment Realities

• Connectivity & Network Limitations: Field staff posted in Sub-Regional Offices (SROs) across tier-2/3 districts may experience unstable 3G/4G bandwidth [KNOWN, INFERENCE]. The platform must optimize frontend assets (<2MB initial load) and support offline caching for self-assessment quizzes with background sync [INFERENCE].
• Hardware Constraints: Official desktop hardware across field offices varies from modern workstations to older PCs running legacy browsers [KNOWN, INFERENCE]. Frontend UI must be responsive, lightweight, cross-browser compatible, and avoid heavy client-side computations [INFERENCE].
• Language & Accessibility: Official statistics work spans English and regional languages across state statistical bureaus [PS, KNOWN]. Assessment generation and virtual assistant interfaces must support English and Hindi initially, with architecture extensible to 22 scheduled languages [INFERENCE].
• Operating Guidelines & Cybersecurity: Deployment must strictly comply with NIC (National Informatics Centre) / CERT-In guidelines, requiring hosting on MeghRaj (Government Cloud) or MeitY-empanelled cloud infrastructure, strict data residency within India, and zero exposure of PII (Personally Identifiable Information) to external proprietary AI APIs [INFERENCE].

6. Research Questions

1. iGOT API Specifications: What are the exact open endpoint standards, authentication handshakes (OAuth2/HMAC), and payload formats supported by Karmayogi Bharat SPV for external LMS integration? [RESEARCHED, NEEDS VALIDATION]
2. NSSTA TPAC Training Structure: How are NSSTA’s Training Programme Advisory Committee (TPAC) annual training calendars structured, and in what digital format are course catalogues published? [RESEARCHED]
3. MoSPI FRAC Framework: Has MoSPI fully finalized its Framework of Roles, Activities, and Competencies (FRAC) dictionary on iGOT Karmayogi, or does the platform need to ingest a custom competency JSON taxonomy? [RESEARCHED, NEEDS VALIDATION]
4. Video Assessment Extraction: What is the optimal open-source speech-to-text pipeline (e.g., Whisper) for processing MoSPI lecture videos with high accuracy for Indian English and Hindi accents? [RESEARCHED]
5. Data Privacy Compliance: What specific data governance policies apply when storing official performance metrics and skill gaps on cloud infrastructure? [KNOWN, INFERENCE]

7. Data Requirements & Sources

8. Existing Solutions & Gap Analysis

9. Candidate Solution Approaches

10. Recommended Approach & Why

The recommended approach is Approach 2: Deep Learning & Vector-Embedding Hybrid System.

Justification & Rejection of Alternatives

• Why Approach 1 (Rule-Based) was Rejected: Rule-based systems cannot scale across the vast diversity of roles in MoSPI (Field Operations Division, National Accounts, Economic Statistics, Price Monitoring) [PS, INFERENCE]. When a new technology like Python or Cloud Computing emerges, every rule must be manually updated [INFERENCE]. It fails the SIH requirement for true AI-driven dynamic recommendations [PS].
• Why Approach 3 (Pure LLM Agent) was Rejected: Relying entirely on LLM agents for deterministic workflows (such as official HR role mapping and scoring) introduces hallucination risks, non-deterministic outputs, extreme token costs, and unacceptable latency during multi-user assessments [INFERENCE].
• Why Approach 2 Wins: Approach 2 combines the determinism of structured relational data (for user roles, target proficiency frameworks, and mathematical gap scoring) with the semantic power of vector embeddings and LLMs (for course similarity matching, multi-format document parsing, and validated MCQ generation with explanations) [INFERENCE]. This guarantees 100% accuracy on official administrative logic while delivering state-of-the-art AI performance for assessment and recommendation [INFERENCE].

11. Technical Architecture

+---------------------------------------------------------------------------------------------------+
|                                      PRESENTATION LAYER                                           |
|  +-----------------------------------+     +---------------------------------------------------+  |
|  |     Learner Dashboard (React)     |     |          Administrator Dashboard (React)          |  |
|  | - Gap Radar Chart                 |     | - Organizational Skill Map                        |  |
|  | - Recommended Pathways            |     | - Training Effectiveness Analytics                |  |
|  | - Interactive Quiz & Assistant    |     | - Predictive Skill Shortage Models                |  |
|  +-----------------------------------+     +---------------------------------------------------+  |
+---------------------------------------------------------------------------------------------------+
                                                 | REST / GraphQL (TLS 1.3)
+---------------------------------------------------------------------------------------------------+
|                                 APPLICATION & INTEGRATION LAYER                                   |
|  +-----------------------------------+     +---------------------------------------------------+  |
|  |    Authentication & RBAC Engine   |     |             iGOT Integration Gateway              |  |
|  |  (OAuth2 / OIDC / Parichay SSO)   |     |  (Catalog Sync, Enrollment, Webhook Listener)     |  |
|  +-----------------------------------+     +---------------------------------------------------+  |
|  +---------------------------------------------------------------------------------------------+  |
|  |                                  API Gateway / Node.js Express                              |  |
|  +---------------------------------------------------------------------------------------------+  |
+---------------------------------------------------------------------------------------------------+
                                                 |
+---------------------------------------------------------------------------------------------------+
|                                      AI & PROCESSING LAYER                                        |
|  +-----------------------------------+     +---------------------------------------------------+  |
|  |   Competency & Gap Engine (Py)    |     |    Intelligent Assessment Engine (LangChain/Py)    |  |
|  | - Delta Matrix Calculator         |     | - Doc/Video Parser (PyMuPDF / Whisper)            |  |
|  | - Embedding Matcher (BGE-m3)      |     | - Constrained MCQ Generator (JSON Schema / LLM)   |  |
|  +-----------------------------------+     +---------------------------------------------------+  |
|  +---------------------------------------------------------------------------------------------+  |
|  |                    RAG Virtual Assistant Engine (ChromaDB / LlamaIndex)                       |  |
|  +---------------------------------------------------------------------------------------------+  |
+---------------------------------------------------------------------------------------------------+
                                                 |
+---------------------------------------------------------------------------------------------------+
|                                   DATA & PERSISTENCE LAYER                                        |
|  +-----------------------------------+     +---------------------------------------------------+  |
|  |   Relational DB (PostgreSQL)      |     |               Vector Store (Qdrant)               |  |
|  | - Users, Roles, FRAC Taxonomy     |     | - Course Embeddings                               |  |
|  | - Quiz Items, Scores, Analytics   |     | - Statistical Document Chunks                     |  |
|  +-----------------------------------+     +---------------------------------------------------+  |
|  +---------------------------------------------------------------------------------------------+  |
|  |                          Object Storage (MinIO / S3 - Courseware PDFs/Videos)                |  |
|  +---------------------------------------------------------------------------------------------+  |
+---------------------------------------------------------------------------------------------------+

End-to-End Data Flow Narrative

1. Ingestion & Profiling: The employee logs in via Single Sign-On (SSO) [PS, INFERENCE]. The platform ingests their profile (Designation: Assistant Director, Division: FOD, Current Skills: SQL, Basic Sampling) [PS, ASSUMPTION].
2. Gap Assessment: The Competency Engine fetches the target profile for "Assistant Director - FOD" from PostgreSQL [INFERENCE]. It calculates the proficiency gap across all 4 domains [PS].
3. Path Recommendation: The Gap Delta vector is queried against Qdrant Vector DB containing embedded course profiles from iGOT Karmayogi and NSSTA TPAC calendars [PS, INFERENCE]. Top-matching courses are ordered by relevance and career priority [PS, INFERENCE].
4. Assessment Generation: A trainer uploads a 20-page NSS Survey Methodology PDF [PS]. The Assessment Engine parses text using PyMuPDF, chunks the text, passes it to the LLM pipeline with a structured JSON output schema, and stores generated MCQs (with distractors and rationale) in PostgreSQL [PS, INFERENCE].
5. Taking Quiz & Feedback: The employee takes the adaptive quiz [PS]. Upon submission, the engine provides instant evaluation, displays explanations for every option, and dynamically updates the learner’s verified competency score in PostgreSQL and syncs with iGOT via API [PS, INFERENCE].
6. Analytics Aggregation: Aggregated progress metrics feed real-time Admin Dashboards, triggering predictive alerts when specific departments show structural skill lags in critical modern domains (e.g., AI/ML or GIS) [PS, INFERENCE].

12. Technology Stack

13. AI/ML Justification

Component 1: Multi-Format Intelligent Assessment Engine (MCQ Generator)

1. Purpose: Automatically extract key statistical concepts from documents/videos and convert them into structured MCQs with rationale [PS].
2. Model Type: LLM (e.g., Llama-3.1-8B / GPT-4o-mini) with Constrained Structured Outputs (Pydantic / JSON Schema) [INFERENCE].
3. Input: Raw PDF text chunks, PPTX slides, or Whisper transcript text [PS, INFERENCE].
4. Output: JSON array of items: { question, options: [A,B,C,D], correct_idx, explanation, difficulty_level, mapped_competency } [PS, INFERENCE].
5. Data Required: Uploaded MoSPI training manuals, survey guides, National Accounts documents [PS, RESEARCHED].
6. Why AI: Manual item creation takes hours per document; standard NLP keyword splitters cannot formulate coherent questions or generate plausible distractors with accurate explanations [INFERENCE].
7. Evaluation: BLEU/ROUGE for ground truth overlap + human SME validation metric (Factuality & Grammatical Correctness > 95%) [INFERENCE].
8. Failure Mode: Model generates ambiguous or factually incorrect questions. Mitigation: Require SME review interface before committing auto-generated questions to live official test banks [INFERENCE].
9. Human Oversight: Mandatory toggle allowing trainers to edit/delete auto-generated questions [PS, INFERENCE].

Component 2: Competency Gap Matcher & Course Recommendation Engine

1. Purpose: Recommend relevant iGOT/NSSTA courses based on employee skill gap vector [PS].
2. Model Type: Dense Vector Embeddings (BGE-m3 or text-embedding-3-small) with Cosine Similarity Matching [INFERENCE].
3. Input: Competency gap delta vector ($Gap = Target - Current$) + User Context [INFERENCE].
4. Output: Ranked list of Course IDs with match confidence percentage and justification [INFERENCE].
5. Data Required: Scraped/Synced course descriptions, syllabus text, and competency tags from iGOT & NSSTA [PS, INFERENCE].
6. Why AI: Keyword matching fails when course titles use different terminology than official competency frameworks (e.g., "Sampling Techniques" vs "Household Survey Design") [INFERENCE].
7. Evaluation: Mean Reciprocal Rank (MRR) and Normalized Discounted Cumulative Gain (NDCG) on test learning paths [INFERENCE].
8. Failure Mode: Recommending overly advanced courses to beginners. Mitigation: Filter vector search results by strict prerequisite level metadata [INFERENCE].
9. Human Oversight: User can provide feedback ("Not relevant") to adjust personal weights [INFERENCE].

Component 3: RAG-Powered Learner Support Assistant

1. Purpose: Provide 24/7 contextual answers to learner queries regarding MoSPI statistical manuals and course concepts [PS].
2. Model Type: Retrieval-Augmented Generation (RAG) using Vector Database + Quantized LLM [INFERENCE].
3. Input: Learner text query + current course context [INFERENCE].
4. Output: Conversational response with direct citation of source document page/section [INFERENCE].
5. Data Required: Ingested MoSPI manuals, statistical definitions, and iGOT course transcripts [PS, RESEARCHED].
6. Why AI: Static FAQs cannot address specific contextual doubts during complex analytical learning [INFERENCE].
7. Evaluation: Ragas framework (Faithfulness, Answer Relevance, Context Recall) [INFERENCE].
8. Failure Mode: Hallucinating statistical formulas or legal rules. Mitigation: Strict system prompts instructing model to state "Information not found in official manual" when context similarity is low [INFERENCE].
9. Human Oversight: Escalation button to submit unresolved doubts to NSSTA faculty [INFERENCE].

Where AI MUST NOT Be Used

• Deterministic User Role Mapping: Assigning mandatory target competencies based on official designation MUST be ruled by deterministic lookup tables, NOT AI guessing [INFERENCE].
• Scoring & Assessment Evaluation: Final percentage calculation and pass/fail state MUST be strictly computed via mathematical functions, NOT LLM evaluation, to prevent non-deterministic grading [INFERENCE].
• Authentication & Authorization: Access control logic MUST be strictly handled by standard security middleware [INFERENCE].

14. Data Architecture & Offline/Low-Network Design

Core Database Schema (PostgreSQL)

-- Users / Officials
CREATE TABLE users (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    official_id VARCHAR(50) UNIQUE NOT NULL,
    full_name VARCHAR(100) NOT NULL,
    email VARCHAR(100) UNIQUE NOT NULL,
    designation VARCHAR(100) NOT NULL,
    department VARCHAR(100) NOT NULL,
    division VARCHAR(100) NOT NULL,
    experience_years INT DEFAULT 0,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

-- Competency Taxonomy Framework (FRAC aligned)
CREATE TABLE competencies (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    domain VARCHAR(50) NOT NULL, -- Statistical, Technical, Digital Governance, Behavioural
    sub_domain VARCHAR(100) NOT NULL,
    competency_name VARCHAR(150) NOT NULL,
    description TEXT,
    max_level INT DEFAULT 5
);

-- User Current Competencies
CREATE TABLE user_competencies (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    user_id UUID REFERENCES users(id) ON DELETE CASCADE,
    competency_id UUID REFERENCES competencies(id),
    current_level INT CHECK (current_level BETWEEN 1 AND 5),
    verified_by_quiz BOOLEAN DEFAULT FALSE,
    updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    UNIQUE(user_id, competency_id)
);

-- Course Catalog (iGOT & NSSTA)
CREATE TABLE courses (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    source_platform VARCHAR(20) NOT NULL, -- 'iGOT' or 'NSSTA_TPAC'
    external_course_id VARCHAR(100),
    title VARCHAR(255) NOT NULL,
    description TEXT,
    duration_hours NUMERIC(5,2),
    course_url VARCHAR(500),
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

-- AI-Generated Assessments
CREATE TABLE assessments (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    title VARCHAR(200) NOT NULL,
    source_document_name VARCHAR(255),
    created_by UUID REFERENCES users(id),
    questions_json JSONB NOT NULL, -- Array of MCQs with options, correct answers, explanations
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Low-Network & Offline Architecture

• Local Caching: Frontend leverages Progressive Web App (PWA) capabilities with IndexedDB caching for downloaded quiz modules and course summary notes [INFERENCE].
• Synchronization & Conflict Handling: When an officer takes a quiz offline in a remote field office, responses are timestamped and signed locally using IndexedDB [INFERENCE]. Upon network reconnection, a background Service Worker pushes responses to the /api/v1/assessments/sync endpoint [INFERENCE]. Conflicts are resolved using Server-Wins rule for competency updates while preserving local completion timestamps [INFERENCE].
• Bandwidth Reduction: Dynamic asset compression (Brotli/Gzip), JSON payload minimization, and lazy loading of dashboard visual elements guarantee functional operation on low-speed 3G field connections [INFERENCE].

15. Security, Privacy & Governance

• Personal & Location Data Protection: Employee PII (email, phone, exact posting station) is segregated from AI processing pipelines [INFERENCE]. Prompt payloads sent to LLMs contain zero direct PII [INFERENCE].
• Authentication & Authorization: Implements OAuth2 / OIDC framework compatible with Parichay (Government Single Sign-On) [INFERENCE]. Strict Role-Based Access Control (RBAC) enforces three levels: Learner, Trainer/SME, and System Admin [INFERENCE].
• Data Encryption: All network traffic uses TLS 1.3 encryption [INFERENCE]. Database storage and object storage Buckets enforce AES-256 encryption at rest [INFERENCE].
• Audit Logging & Compliance: Immutable audit log table records all profile edits, assessment generation triggers, competency manual overrides, and system administrative actions [INFERENCE]. System design strictly aligns with CERT-In security guidelines and the Digital Personal Data Protection (DPDP) Act 2023 [KNOWN, INFERENCE].
• AI Model Governance & Misuse Prevention: Auto-generated MCQs MUST be marked as "Draft / AI-Generated" until explicitly approved or verified by a qualified SME before being used in formal promotion or official evaluation tracks [INFERENCE].

16. MVP Scope (Hackathon Build)

+----------------------------------------------------------------------------------------------------+
|                                    HACKATHON MVP SCOPE                                             |
+----------------------------------------------------------------------------------------------------+
| MUST BUILD (Core Capabilities)                                                                     |
|  - Mock User Profiling & HR Data Import (ISS/SSS/FOD Role Archetypes)                             |
|  - FRAC-Aligned Competency Matrix Store across 4 Domains                                          |
|  - Skill Gap Identification Engine ($Gap = Target - Current$)                                     |
|  - Hybrid Vector Recommendation Engine for iGOT & NSSTA TPAC Courses                               |
|  - Multi-Format Document Parser (PDF/PPTX) & LLM MCQ Generation with Rationale                     |
|  - Interactive Learner Dashboard (Gap Radar Chart, Course Path, Instant Quiz Player)              |
|  - Administrator Dashboard (Skill Distribution Heatmap, Aggregated Gaps)                         |
|  - Mock iGOT API Synchronization Bridge & Webhook Endpoints                                       |
+----------------------------------------------------------------------------------------------------+
| SHOULD BUILD (Differentiators)                                                                     |
|  - RAG-based AI Virtual Assistant for statistical documentation queries                            |
|  - Video Transcript Ingestion (Whisper integration for video URL / file upload)                   |
|  - Export Assessment to PDF / QTI / CSV format for trainers                                        |
+----------------------------------------------------------------------------------------------------+
| NICE TO HAVE (If Time Permits)                                                                     |
|  - Multilingual Toggle (English / Hindi UI & Quiz generation)                                     |
|  - Offline PWA Service Worker for local quiz attempts                                              |
+----------------------------------------------------------------------------------------------------+
| DO NOT BUILD DURING HACKATHON (Out of Scope)                                                      |
|  - Live Integration with Production NIC Parichay Servers (Use Mock OAuth OIDC)                     |
|  - Custom Native Mobile Apps (iOS/Android) (Web-responsive UI is sufficient)                       |
|  - Full Virtual Laboratory Cloud Containers (Provide simulated links instead)                      |
+----------------------------------------------------------------------------------------------------+

17. Live Demo Simulation Plan

To present a flawless, high-impact demonstration during SIH judging without relying on closed government production APIs, the team will execute a Transparent Simulation Architecture:

1. Simulated iGOT Karmayogi API Bridge:
   • Real-World Barrier: Production iGOT APIs require official Karmayogi Bharat OAuth credentials and whitelist access [RESEARCHED].
   • Simulation Strategy: Build a mock Express API server implementing exact iGOT API endpoints (/api/v1/igot/catalog, /api/v1/igot/enrollment, /api/v1/igot/competency-update) returning real scraped metadata from 100+ statistical courses [ASSUMPTION, INFERENCE].
2. Live Document Ingestion & MCQ Generation:
   • Demo Walkthrough: Live upload an actual MoSPI publication PDF (e.g., NSS 78th Round Key Indicators Document) during the presentation [ASSUMPTION].
   • Execution: Show real-time extraction, prompt pipeline execution, and immediate rendering of 5 structured MCQs with detailed answer rationales [INFERENCE].
3. Role-Based Journey Demonstration:
   • Persona A (Learner - Field Officer): Log in as a Field Investigator -> View skill gap radar chart -> See recommended Python & Survey Sampling iGOT courses -> Take generated quiz -> Watch gap radar update in real time [INFERENCE].
   • Persona B (Administrator / DIID Head): Log in as Admin -> View national competency heatmap -> Observe predictive lag alert in "GIS Analytics" -> Click "Trigger Recommended Capacity Program" [INFERENCE].

18. Implementation Roadmap

Target Deadline: 30 September 2026 [PS]

Phase 1: Architecture & Data Modeling (Months 1 - 2)
[Milestone 1: DB Schemas, FRAC Taxonomy JSON, Mock Data Generation]
 ├── Design PostgreSQL & Vector schemas
 ├── Curate MoSPI Competency Dictionary across 4 domains
 └── Build Mock iGOT Catalog & User Profile dataset

Phase 2: Core Engine & AI Development (Months 3 - 5)
[Milestone 2: Gap Engine, Vector Recommendation, MCQ Pipeline]
 ├── Build Skill Gap Matrix Engine ($Gap = Target - Current$)
 ├── Implement Qdrant Embedding Pipeline for course matching
 └── Build PyMuPDF/Whisper + LLM MCQ Generator with JSON enforcer

Phase 3: Web Application & Dashboards (Months 6 - 8)
[Milestone 3: React Learner & Admin Dashboards, RAG Assistant]
 ├── Develop responsive Next.js frontend with Shadcn UI & Recharts
 ├── Implement RAG-based AI Virtual Support Assistant
 └── Build Trainer Document Upload & Quiz Editing UI

Phase 4: Integration, Testing & Security (Months 9 - 11)
[Milestone 4: iGOT Mock API Bridge, Role-Based Security, Testing]
 ├── Integrate mock iGOT API sync and webhook listeners
 ├── Execute CERT-In style vulnerability testing & OWASP checks
 └── Perform low-bandwidth offline sync validation

Phase 5: Refinement, SIH Demo Prep & Final Deployment (Month 12 -> Sept 30, 2026)
[Milestone 5: Production Dockerization, Live Demo Script, Final Dossier]
 ├── Final end-to-end containerization (Docker Compose)
 ├── Prepare live judging demo environment with pre-loaded datasets
 └── Project Code Freeze and Final Presentation Submission [30 Sept 2026]

19. Testing & Validation Plan

• Functional Testing: 100% test coverage on API endpoints (Gap calculation, Auth, Quiz evaluation) using Jest and PyTest [INFERENCE].
• AI Model & MCQ Quality Validation:
  • Test Dataset: Run 50 standard MoSPI statistical PDFs through the MCQ Generator [ASSUMPTION].
  • Metrics: Verify zero JSON schema parsing errors; evaluate question accuracy with a sample of statistical experts (Target: >92% factual accuracy) [INFERENCE].
  • False Positive / Negative Testing: Ensure distractors in MCQs are distinct and unambiguously incorrect; verify correct answer logic [INFERENCE].
• Performance Testing: Use Apache JMeter to simulate 500 concurrent learners taking quizzes simultaneously; ensure API response latency remains <200ms and vector query response <100ms [INFERENCE].
• Security Testing: OWASP Top 10 vulnerability scanning, SQL injection checks, JWT tamper testing, and PII leakage audit on LLM prompt outputs [INFERENCE].
• Offline & Sync Validation: Simulate network disconnection during quiz attempt; verify local IndexedDB store and successful auto-sync upon network restoration [INFERENCE].

20. Risk Register

21. Cost & Scalability

Hackathon Phase

• Compute & Hosting: $0 (Local development environment + Free tier cloud deployment on Vercel / Render / Railway) [INFERENCE].
• AI API Cost: Minimal (< $15 total using OpenAI API gpt-4o-mini credits or local free Ollama Llama-3.1-8B models) [INFERENCE].
• Database: Free tier PostgreSQL (Neon / Supabase) + Local Qdrant / ChromaDB container [INFERENCE].

Production Deployment (MoSPI / DIID Enterprise Scale - ~30,000 Officials)

• Cloud Infrastructure: NIC Cloud (MeghRaj) virtual machines with GPU instance (e.g., 1x NVIDIA A10G / L4 for local open-source LLM inference) [KNOWN, INFERENCE].
• Database & Storage: Managed PostgreSQL Cluster with read-replicas + Self-hosted Qdrant Cluster + Object Storage (MinIO) [INFERENCE].
• Scalability Design: Horizontal auto-scaling for API gateway; stateless web tiers; vector index caching in Redis; asynchronous queuing for document processing pipelines [INFERENCE].

22. Differentiation

23. Future Scope

1. AI Virtual Laboratories Integration: Provide embedded cloud-based Jupyter Notebook sandbox environments for hands-on Python, R, and SQL exercises directly within the LMS [PS, INFERENCE].
2. Voice-Based Native Language Assistant: Expand AI Virtual Assistant to support voice interaction across 22 Indian official languages using Bhashini APIs [INFERENCE].
3. Career Progression Predictor: Leverage machine learning to recommend non-linear career learning paths aligned with upcoming Indian Statistical Service (ISS) promotions and postings [PS, INFERENCE].
4. Automatic Certificate Verification via India Stack: Issue Verifiable Credentials (VCs) on DigiLocker upon course and assessment completion [INFERENCE].

24. Assumptions & Open Items Log

Consolidated Assumptions

• [ASSUMPTION-01]: MoSPI HRMS data can be represented via a structured JSON schema containing designation, department, and past training history.
• [ASSUMPTION-02]: A scraped/synthetic course catalog of 100+ statistical courses accurately reflects iGOT Karmayogi’s current offerings for hackathon demonstration.
• [ASSUMPTION-03]: Open-source LLMs (Llama-3.1-8B) and embeddings (BGE-m3) provide sufficient accuracy for statistical document parsing and question generation.

Consolidated Items Needing Validation

• [NEEDS VALIDATION-01]: Official production REST API specification and developer keys from Karmayogi Bharat SPV for live iGOT integration.
• [NEEDS VALIDATION-02]: Finalized FRAC dictionary mappings specifically approved by MoSPI and the Capacity Building Commission (CBC).
• [NEEDS VALIDATION-03]: Exact deployment security policies required by MoSPI DIID for cloud hosting on MeghRaj.

Final Recommendation

Adopt Approach 2 (Deep Learning & Vector-Embedding Hybrid Architecture). It satisfies all explicit MoSPI requirements, ensures zero non-deterministic errors in official administrative mapping, provides state-of-the-art AI assessment generation from complex statistical materials, and delivers an enterprise-ready architecture for iGOT Karmayogi integration [PS, INFERENCE].

Recommended Stack

• Frontend: React / Next.js 14, Tailwind CSS, Shadcn UI, Recharts [INFERENCE].
• Backend: Node.js (Express Gateway) + Python FastAPI (AI Microservices) [INFERENCE].
• Databases: PostgreSQL 16 (Relational/FRAC) + Qdrant (Vector Store) + Redis (Cache) [INFERENCE].
• AI Engine: LangChain, PyMuPDF, OpenAI Whisper, BGE-m3 Embeddings, Ollama / Llama-3.1-8B / GPT-4o-mini [INFERENCE].

Biggest Dependency

Access to structured iGOT Karmayogi course metadata and MoSPI’s official FRAC competency taxonomy dictionary [PS, RESEARCHED].

Biggest Risk

Potential LLM hallucinations or factual inaccuracies when generating MCQs from highly complex mathematical/statistical manuals [INFERENCE].

Biggest Technical Challenge

Engineered prompt pipelines with constrained JSON schema enforcement to guarantee 100% syntactically valid and factually correct question-answer-rationale output from multi-format document/video sources [INFERENCE].

First 5 Actions

1. Define the PostgreSQL database schema for Users, FRAC Competency Taxonomy, and iGOT Course Catalog [INFERENCE].
2. Curate a structured JSON dataset representing MoSPI’s 4 core competency domains and 100+ sample statistical courses [INFERENCE].
3. Build the core Skill Gap Calculation Engine ($Gap = Target - Current$) in Python [INFERENCE].
4. Develop the multi-format document parser (PyMuPDF) and LLM MCQ Generation pipeline with constrained JSON output [INFERENCE].
5. Construct the Next.js Learner Dashboard featuring the interactive skill gap radar chart and recommendation feed [INFERENCE].

Validation Before Coding

• Verify PyMuPDF extraction accuracy on a multi-column MoSPI statistical report PDF [INFERENCE].
• Test LLM prompt templates against 5 complex statistical paragraphs to ensure distractors and rationales are 100% accurate [INFERENCE].

Buildability Score /100

92 / 100 (Extremely buildable by a 4–6 student team within the SIH timeframe using standard modern web and open AI frameworks) [INFERENCE].

Confidence

High [PS, RESEARCHED, INFERENCE] — Clear source of truth requirements, well-defined architecture, and fully verifiable technology stack.

Research Sources

1. Title: Press Information Bureau - iGOT Karmayogi Platform Capabilities & Reach
   • URL: https://www.pib.gov.in/PressReleasePage.aspx?PRID=2223638&reg=3&lang=1 [1.9]
   • Source Type: Official Government Source (PIB)
   • Supported Claim: Documents >1.48 Crore users, 4200+ courses, Mission Karmayogi scale, and multi-lingual support.
   • Reliability: High / Official
2. Title: Capacity Development Division - National Statistical Systems Training Academy (NSSTA) MoSPI
   • URL: https://www.mospi.gov.in/about-us/nssta [2.1]
   • Source Type: Official Government Portal (mospi.gov.in)
   • Supported Claim: Confirms NSSTA mandate, TPAC training calendar structure, and capacity development functions under MoSPI.
   • Reliability: High / Official
3. Title: Karmyogi Competency Model (KCM) - Capacity Building Commission
   • URL: https://cbc.gov.in/karmyogi-competency-model-kcm [3.1]
   • Source Type: Official Government Portal (cbc.gov.in)
   • Supported Claim: Validates FRAC framework structure, roles, activities, and competency mapping standards.
   • Reliability: High / Official
4. Title: Data Informatics & Innovation Division (DIID) - Data Innovation Lab Initiative
   • URL: https://www.pib.gov.in/PressNoteDetails.aspx?NoteId=157886&ModuleId=3&reg=22&lang=13 [4.8]
   • Source Type: Official Government Source (PIB)
   • Supported Claim: Highlights MoSPI DIID's initiative to integrate AI/ML, semantic search, and modern technological capacity into the Indian Statistical System.
   • Reliability: High / Official