BharatGov AI

India's AI-powered government-services assistant

BharatGov AI is a privacy-conscious assistant for Indian government services. Ask questions in plain language about schemes, documents and official procedures — it answers with citations drawn from a local knowledge base of official government sources, so you know exactly where the information came from. The app is multilingual-ready and works even when paid AI credits run out.

Try it live: https://bharatgov-ai-phi.vercel.app

---

Features

• Ask about any government service — schemes, eligibility, required documents, application steps, helplines and more.
• Source-aware answers with citations — every answer is grounded in a curated knowledge base of official government documents and returns clickable sources.
• Multi-provider AI with automatic fallback — answers are generated by Gemini, then OpenAI, and finally a live web-search summary — the app never stops working.
• Document and category browser — explore the underlying knowledge base by category (/categories, /documents).
• Account dashboard with chat history — register, log in, and review your past conversations.
• Privacy-conscious by design — chat history is stored per account and the assistant is engineered to avoid collecting unnecessary personal data.

Tech Stack

Layer	Technology
Web app	Next.js 15 (App Router), TypeScript, Tailwind CSS
API	Python 3.12, FastAPI, SQLAlchemy, Pydantic
AI providers	Google Gemini REST, OpenAI Responses API, DuckDuckGo web search (fallback)
Storage	SQLite (local / single instance), PostgreSQL + Redis + Qdrant (Docker stack)
Auth	Email/password (bcrypt + JWT), optional Google OAuth
Deployment	Web on Vercel, API on Railway

Architecture

┌──────────────────────────┐          ┌─────────────────────────────┐
│  Next.js web app          │          │  FastAPI backend            │
│  apps/web                 │ ───────► │  services/api               │
│  - Login / Signup / Chat  │  /api/v1 │  - Auth (JWT)               │
│  - Dashboard              │          │  - /ask, /chat, /documents  │
│  - Vercel (vercel.app)    │          │  - /categories, /feedback   │
└──────────────────────────┘          │  - Railway (railway.app)    │
                                      └─────────────┬───────────────┘
                                                    │
                                     ┌──────────────┼──────────────┐
                                     ▼              ▼              ▼
                                Gemini 2.0     OpenAI GPT-4o-mini  DuckDuckGo
                               (primary LLM)   (fallback LLM)    (last-resort search)

API Overview

Base URL: https://bharatgov-ai-production.up.railway.app/api/v1 — interactive docs at /docs.

Endpoint	Method	Description
/auth/register	POST	Create an account
/auth/login	POST	Log in, returns a JWT
/auth/me	GET	Current user profile
/ask	POST	Ask the assistant (grounded in the knowledge base)
/chat	GET / POST	List / create conversations
/chat/{chat_id}	GET	Fetch a conversation's messages
/categories	GET	Knowledge-base categories
/documents	GET	Government source documents
/feedback	POST	Submit answer feedback

Getting Started (local, no Docker)

# 1. Environment
cp .env.example .env

# 2. API (port 8000)
cd services/api
python3 -m venv .venv && source .venv/bin/activate
pip install -e '.[dev]'
python -m app.db.seed          # loads categories + 39 government source documents
uvicorn app.main:app --reload

# 3. Web (port 3000) — in a second terminal
cd apps/web
npm install
npm run dev

Open http://localhost:3000. API docs at http://localhost:8000/docs.

Docker

cp .env.example .env
docker compose up --build

The Docker stack uses PostgreSQL, Redis and Qdrant. Run the seed step once against the container DB (docker exec <api-container> python -m app.db.seed).

Environment Variables

Variable	Used by	Required
JWT_SECRET	API	Yes
OPENAI_API_KEY	API (fallback LLM)	No
GEMINI_API_KEY	API (primary LLM)	No
CORS_ORIGINS	API	Yes (deployed origins)
NEXT_PUBLIC_API_URL	Web	No (auto-resolved at runtime)
GOOGLE_CLIENT_ID / GOOGLE_CLIENT_SECRET	API	Optional
NEXT_PUBLIC_GOOGLE_CLIENT_ID	Web	Optional

Deployment

• Web: deployed on Vercel at https://bharatgov-ai-phi.vercel.app. The build is framework-native (no Docker needed) — apps/web runs next build.
• API: deployed on Railway at https://bharatgov-ai-production.up.railway.app. The API is the single source of truth for auth and answers; the web app points to it via NEXT_PUBLIC_API_URL.
• Every CORS_ORIGINS entry must list each deployed web origin.

See docs/DEVELOPMENT.md and docs/DEPLOYMENT.md for details.

Project Structure

.
├── apps/
│   └── web/                 # Next.js frontend (login, chat, dashboard)
├── services/
│   └── api/                 # FastAPI backend
│       ├── app/api/routes/  # auth, chat, resources
│       ├── app/services/    # AI assistant (Gemini → OpenAI → web search)
│       ├── app/db/          # models + seed data
│       └── app/core/        # config, security, settings
├── docs/
├── docker-compose.yml
└── .env.example

License

MIT