SchemeAI — Government Scheme Matcher for Entrepreneurs

Built for Smart India Hackathon 2026 · Problem Statement #26092

SchemeAI is a web app that helps marginalized entrepreneurs — women, SC/ST/OBC, minorities, and PwD — discover and apply for relevant government schemes. It uses AI-powered matching to cut through the complexity of 1,000+ government schemes and show only what you're actually eligible for.

---

Features

• AI Scheme Matching — Rule-based hard filtering combined with Sentence-BERT semantic matching ranks schemes by relevance to your business description
• 10 Regional Languages — Full UI and scheme summaries in English, Hindi, Marathi, Tamil, Telugu, Bengali, Gujarati, Kannada, Malayalam, and Punjabi
• Voice Input & Output — Speak your business details using the microphone; listen to any scheme summary in your selected language
• AI Chatbot Assistant — Ask the chatbot about any scheme, eligibility, documents, or how to apply — answers in your chosen language
• Live Scheme Data — Schemes are automatically fetched from myscheme.gov.in every 6 hours so data stays current
• Instant Results — Matching takes under 2 seconds; embeddings are pre-cached locally

---

Tech Stack

Layer	Technology
Frontend	React 18, Tailwind CSS, Vite
Backend	FastAPI (Python), Uvicorn
AI Matching	sentence-transformers (all-MiniLM-L6-v2), TF-IDF fallback
Translation	Google Translate API (GTX endpoint)
TTS	Browser Web Speech API (primary), gTTS (fallback)
Live Data	myscheme.gov.in public API, auto-synced every 6h

---

How It Works

User fills 5-step profile (or speaks it)
            │
    Rule Engine (hard filters)
    Category, State, Stage, Income
            │
    Semantic Engine (Sentence-BERT)
    Cosine similarity on business description
            │
    Combined Score = (Rule × 0.6) + (Semantic × 0.4)
            │
    Ranked scheme cards with summaries in your language

---

Quick Start

Windows (one-click)

Double-click run_app.bat — opens the app at http://localhost:8080 automatically.

Manual

# Backend
cd backend
pip install -r requirements.txt
python -m uvicorn main:app --host 0.0.0.0 --port 8080

The backend also serves the pre-built frontend, so no separate frontend server is needed.

To rebuild the frontend after making changes:

cd frontend
npm install
npm run build

---

Project Structure

├── backend/
│   ├── main.py               # FastAPI app, background auto-sync
│   ├── data/
│   │   └── schemes.json      # Curated base scheme database
│   ├── matching/
│   │   ├── nlp_engine.py     # Sentence-BERT + TF-IDF matching
│   │   ├── rule_engine.py    # Hard eligibility filters
│   │   └── live_fetcher.py   # myscheme.gov.in auto-sync engine
│   ├── routers/
│   │   ├── schemes.py        # Matching, live search, sync endpoints
│   │   ├── chatbot.py        # Multilingual AI chatbot
│   │   ├── tts.py            # Text-to-speech endpoint
│   │   └── profile.py        # User profile management
│   └── utils/
│       └── translation.py    # 10-language translation engine
├── frontend/
│   └── src/
│       ├── pages/            # LandingPage, MatchPage
│       └── components/       # SchemeCard, SchemeChatbot, VoiceOutput, ...
├── run_app.bat               # One-click Windows launcher
└── start_servers.bat         # Dev mode launcher

---

API Endpoints

Method	Endpoint	Description
POST	/api/profile	Create user profile
GET	/api/schemes/match?profile_id=	Get AI-matched schemes
GET	/api/schemes/live-search?q=	Search live schemes
POST	/api/schemes/sync-live	Trigger immediate sync
GET	/api/schemes/{id}/summary?lang=hi	Get localized scheme summary
POST	/api/chat	Send message to AI chatbot
POST	/api/tts	Convert text to speech
GET	/api/health	Health check

---

License

MIT