跳到正文
GitHub·· 5 天前AI 评分15

SchemeSaathi：AI 驱动的印度政府福利方案导航与确定性资格引擎

nikhi20-900/SchemeSaathi

AI 导读

SchemeSaathi 是面向印度公民的 AI 政府福利方案导航器，结合自然语言理解、OCR 文档智能与确定性规则引擎，按官方标准评估福利资格，避免 AI 幻觉。资格引擎已实现并通过 48 项测试，前端 9 页 UI 与交互演示已完成并合并；AI/RAG 管线与 OCR 文档智能仍在开发中。

正文

AI-Powered Government Scheme Navigator & Deterministic Eligibility Engine

SchemeSaathi is an AI-powered Government Scheme Navigator for Indian citizens, combining natural language comprehension, OCR-driven document intelligence, and a deterministic rules engine to evaluate welfare eligibility against published government criteria with zero AI hallucinations.

PROJECT STATUS: PHASE 2 (CORE MODULE IMPLEMENTATION)

  • ✅ Member 2 (Eligibility Engine): Fully implemented, tested (48 tests), merged to main via PR #6.
  • ✅ Member 4 (Frontend & Integration): Complete 9-page civic-tech UI, interactive demo walkthrough, merged to main via PR #7.
  • 🔄 Member 1 (AI / RAG Pipeline): In Progress (Issue #1).
  • 🔄 Member 3 (OCR / Document Intelligence): In Progress (Issue #3).

🔒 Core Architectural Principle

┌────────────────────────────────────────────────────────────────────────┐
│  🔒 THE LLM DOES NOT INDEPENDENTLY DETERMINE CITIZEN ELIGIBILITY.       │
│                                                                        │
│  Eligibility criteria (income limits, age, state domicile, categories) │
│  are evaluated deterministically by the Rules Engine against published │
│  official government rules. The LLM only handles natural language      │
│  understanding, semantic search, and grounded multilingual explanation.│
└────────────────────────────────────────────────────────────────────────┘

👥 4-Member Team Ownership & Status

                    MEMBER 1 🔄
                 AI + RAG + DATA
                        │
               ┌────────┴────────┐
               ▼                 ▼
          MEMBER 2 ✅        MEMBER 3 🔄
        ELIGIBILITY              OCR
          ENGINE              DOCUMENTS
               │                 │
               └────────┬────────┘
                        ▼
                    MEMBER 4 ✅
             FRONTEND + INTEGRATION
Member Focus Area Branch Status Key Deliverables
Member 1 AI + RAG + Data feature/ai-rag 🔄 In Progress Vector embeddings, hybrid retrieval, grounding in official gazettes (Issue #1)
Member 2 Eligibility Engine feature/eligibility-engine ✅ Completed & Merged Deterministic rules engine, Pydantic schemas, 8 operators, 48 tests (PR #6)
Member 3 OCR + Documents feature/ocr-documents 🔄 In Progress Document parsing, certificate field extraction, fraud detection (Issue #3)
Member 4 Frontend + Integration feature/frontend ✅ Completed & Merged 9 core pages, interactive demo workflow, client-side fallback engine (PR #7)

✨ Features Implemented Till Now

1. Deterministic Eligibility Engine (Member 2)

  • Engine Core (backend/app/eligibility/engine.py):
    • Strictly evaluates profile fields against rules without AI ambiguity.
    • Multi-tier verdicts:
      • PASS: Profile value satisfies the threshold.
      • FAIL: Profile value violates the threshold.
      • NEEDS_VERIFICATION: Required profile attribute or document certificate is unverified or missing.
    • Overall scheme evaluation: ELIGIBLE, PARTIALLY_ELIGIBLE, INELIGIBLE.
  • Operator Library (backend/app/eligibility/operators.py):
    • Numeric comparisons: >=, <=, >, <
    • Equality checks: equals, ==, not_equals, !=
    • Set memberships: in, not_in, contains
  • FastAPI Endpoints (backend/app/api/eligibility.py):
    • POST /api/eligibility/check: Evaluate profile against a single scheme or all registered schemes.
    • GET /api/eligibility/schemes: Return catalog of schemes with structured rules.
    • GET /api/eligibility/schemes/{scheme_id}: Retrieve rules for a specific scheme.
  • Reference Scheme Registry (backend/app/eligibility/scheme_rules.py):
    • Karnataka Post-Matric Scholarship for OBC Students (KA-SCHOLARSHIP-001)
    • Pradhan Mantri Kaushal Vikas Yojana (IN-PMKVY-001)
    • Pradhan Mantri Awas Yojana – Gramin (IN-PMAY-001)
    • Karnataka Raitha Siri Scheme (KA-FARM-001)
  • Comprehensive Test Suite (backend/tests/test_eligibility.py):
    • 48 test cases verifying all operators, edge cases, partial matches, and schema validations.

2. Frontend Application & Civic Design System (Member 4)

  • Aesthetic: Clean, accessible civic-tech design system built with Tailwind CSS, Framer Motion, and Lucide React.
    • Curated palette: Charcoal (#151B18), Linen (#FAF9F6), Terracotta (#C85A32), Sage (#3F6152), and Saffron (#D97706).
  • 9 Core Application Views:
    1. Landing / Home (pages/Home.tsx): Value proposition, 4-step workflow, and instant demo entrypoint.
    2. Citizen Profile (pages/Profile.tsx): Form capturing age, state (28 states dropdown), district, education, category, student enrollment, and annual income.
    3. Schemes Directory (pages/Schemes.tsx): Filterable directory with live search and state-level filters.
    4. Scheme Details (pages/SchemeDetails.tsx): Benefits summary, structured eligibility criteria table, required documents checklist, and official portal links.
    5. AI Scheme Navigator (pages/Assistant.tsx): Natural language chat interface with suggested inquiries, grounded gazette citations, and action triggers.
    6. Eligibility Results Dashboard (pages/Results.tsx): Summary cards, filter tabs, and expandable criteria tables with distinct PASS, FAIL, and NEEDS_VERIFICATION badges.
    7. Document Intelligence & OCR (pages/Documents.tsx): File upload dropzone, 4 pre-loaded 1-click test certificates, multi-stage OCR animation, extracted key-value comparison table with confidence scores, and profile update mechanism.
    8. Official Evidence Registry (pages/Evidence.tsx): Auditable gazette notification orders (G.O.), verbatim excerpts, and SHA-256 cryptographic hashes.
    9. Root Navigation & Demo Stepper (App.tsx): Persistent top bar guiding evaluators through the 6-stage verification flow with a 1-click reset button.
  • Client-Side Engine Fallback (frontend/src/utils/engine.ts):
    • Pure TypeScript implementation of the rules engine ensuring 100% demo reliability even if the backend is offline.

🔄 End-to-End Demo Workflow

Landing (Home)
  ↓
Citizen Profile Setup (Nikhil Kumar, Age 20, Karnataka, Category 2A, ₹2,40,000/yr)
  ↓
AI Scheme Navigator (Ask questions & receive grounded citations)
  ↓
Browse Schemes (Karnataka Post-Matric OBC Scholarship)
  ↓
Eligibility Results (Initial evaluation: Partially Eligible — Domicile Certificate missing)
  ↓
Document Intelligence (Select Karnataka Domicile Certificate demo sample)
  ↓
OCR Verification (Extracts name, state, and >7 yrs residence with 98% confidence)
  ↓
Apply to Profile & Re-evaluate
  ↓
Eligibility Results (100% Eligible — PASS with all 8 criteria satisfied)
  ↓
Official Gazette Evidence View (Auditable legal proof & notification orders)

📁 Repository Structure

schemesaathi/
│
├── README.md                           # Project documentation & team progress
├── .gitignore                          # Git ignore rules
├── .env.example                        # Environment variables template
├── docker-compose.yml                  # Multi-container orchestration (FastAPI + Vite + Postgres)
├── .vscode/
│   └── settings.json                   # IDE CSS linter rules for Tailwind directives
│
├── frontend/                           # React + TypeScript + Vite + Tailwind CSS
│   ├── package.json
│   ├── vite.config.ts
│   ├── tsconfig.json
│   ├── index.html
│   ├── README.md                       # Frontend architecture & design guide
│   │
│   └── src/
│       ├── main.tsx
│       ├── App.tsx                     # Main layout & Hackathon demo walkthrough stepper
│       ├── index.css                   # Civic design system tokens & utilities
│       │
│       ├── components/
│       │   ├── Navbar.tsx              # Top sticky navigation bar
│       │   ├── SchemeCard.tsx          # Card displaying scheme overview & criteria count
│       │   ├── EligibilityBadge.tsx    # PASS / FAIL / NEEDS_VERIFICATION badges
│       │   ├── DocumentCard.tsx        # Document preview & upload cards
│       │   ├── EvidenceCard.tsx        # Gazette evidence citation card
│       │   └── LoadingState.tsx        # Animated loading spinners & pipeline steppers
│       │
│       ├── pages/
│       │   ├── Home.tsx                # 1. Landing page
│       │   ├── Profile.tsx             # 2. Citizen profile form
│       │   ├── Schemes.tsx             # 3. Searchable schemes directory
│       │   ├── SchemeDetails.tsx       # 4. Detailed scheme breakdown
│       │   ├── Assistant.tsx           # 5. Grounded AI scheme navigator
│       │   ├── Results.tsx             # 6. Deterministic eligibility results dashboard
│       │   ├── Documents.tsx           # 7 & 8. Document upload & OCR verification
│       │   └── Evidence.tsx            # 9. Gazette orders & audit registry
│       │
│       ├── services/
│       │   └── api.ts                  # Typed API client with mock fallback data
│       ├── types/
│       │   └── index.ts                # TypeScript domain models & interfaces
│       └── utils/
│           └── engine.ts               # Client-side deterministic evaluation engine
│
├── backend/                            # FastAPI + Python 3.11 + Pydantic v2
│   ├── requirements.txt
│   ├── Dockerfile
│   ├── .dockerignore
│   │
│   ├── app/
│   │   ├── __init__.py
│   │   ├── main.py                     # FastAPI application entrypoint & CORS
│   │   │
│   │   ├── api/
│   │   │   ├── __init__.py
│   │   │   ├── eligibility.py          # Eligibility check & scheme endpoints (Active)
│   │   │   ├── profile.py              # Profile endpoints
│   │   │   ├── schemes.py              # Schemes endpoints
│   │   │   ├── assistant.py            # AI assistant endpoints
│   │   │   ├── documents.py            # Document upload & verify endpoints
│   │   │   └── demo.py                 # Demo endpoints
│   │   │
│   │   ├── eligibility/                # Deterministic rules engine (Member 2 - Complete)
│   │   │   ├── __init__.py
│   │   │   ├── engine.py               # Core evaluation algorithm
│   │   │   ├── operators.py            # Comparison operators (>=, <=, in, equals)
│   │   │   ├── validators.py           # Input schema validation
│   │   │   └── scheme_rules.py         # Sample government scheme definitions
│   │   │
│   │   ├── schemas/
│   │   │   ├── eligibility.py          # Pydantic v2 validation models
│   │   │   └── ...
│   │   │
│   │   ├── services/
│   │   │   ├── eligibility_service.py  # Orchestrates profile evaluation against rules
│   │   │   └── ...
│   │   │
│   │   └── core/
│   │       └── config.py               # Pydantic BaseSettings
│   │
│   └── tests/
│       ├── test_eligibility.py         # 48 eligibility engine unit tests (All passing)
│       ├── test_rag.py                 # RAG pipeline tests
│       └── test_documents.py           # Document tests
│
├── data/
│   ├── schemes/                        # Raw gazette documents & policy PDFs
│   └── sample/                         # Sample citizen certificates
│
└── docs/
    ├── architecture.md                 # System architecture specification
    ├── api.md                          # API contract documentation
    └── demo.md                         # Hackathon demonstration script

🚀 Getting Started

Prerequisites

  • Node.js (v18+) & npm
  • Python (3.11+)
  • Docker & Docker Compose (optional)

1. Running the Frontend

cd frontend
npm install
npm run dev

Open http://localhost:5173 in your browser.

To run the TypeScript build check:

npm run build

2. Running the Backend

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Open API Docs at http://localhost:8000/docs.

3. Running Backend Tests

cd backend
pytest

(All 53 unit and integration tests execute in <0.3s)

4. Running with Docker Compose

docker-compose up --build

🚦 Roadmap & Progress Checklist

  • Phase 1: Project Scaffolding

    • Repository structure, branch strategy, and CI standards established
    • Initial FastAPI routing and React UI scaffolding configured
    • PR #5 merged to main

  • Phase 2: Core Engineering Modules

    • Member 2: Deterministic Eligibility Engine (Merged in PR #6)
      • Rule evaluation engine (engine.py)
      • Operator library (operators.py)
      • Pydantic validation schemas (schemas/eligibility.py)
      • API endpoints (/api/eligibility/check, /api/eligibility/schemes)
      • 48 unit tests passing
    • Member 4: Frontend UI & Guided Demo Flow (Merged in PR #7)
      • 9 core pages and civic-tech design system
      • Interactive 6-step demo walkthrough stepper
      • Document OCR upload & verification preview
      • Grounded AI Scheme Navigator with gazette citations
      • Client-side deterministic evaluation fallback
      • Clean TypeScript build (npm run build)
    • Member 1: AI / RAG Pipeline (Issue #1)
      • Gazette text extraction and chunking
      • Vector database indexing and hybrid retrieval
      • Grounded multilingual synthesis
    • Member 3: Document Intelligence & OCR Pipeline (Issue #3)
      • Tesseract / PaddleOCR image preprocessing
      • Certificate field extraction & entity parsing
      • Mismatch detection & fraud prevention logic

  • Phase 3: Final Integration & Hackathon Polish

    • Connect live RAG pipeline to Frontend AI Navigator
    • Connect live OCR service to Document Upload view
    • End-to-end multi-container demo deployment

来源：GitHub · github.com

Agent 智能体开源生态产品更新#产品更新#RAG#开源/仓库#Agent