An end-to-end Adaptive Retrieval-Augmented Generation (RAG) system with a full authentication layer. Ask questions against your uploaded documents — the system intelligently decides whether to search your knowledge base, answer from general knowledge, or run a live web search.
┌─────────────────────────────────────────────────────────────┐
│ Streamlit Frontend │
│ ┌──────────────┐ ┌──────────────────────────────┐ │
│ │ Login Page │ ────► │ Chat UI + Document Upload │ │
│ └──────────────┘ └──────────────────────────────┘ │
└──────────────────────────────┬──────────────────────────────┘
│ JWT Bearer Token
▼
┌─────────────────────────────────────────────────────────────┐
│ Auth Service (FastAPI :8080) │
│ POST /auth/register · POST /auth/login · GET /user/me │
│ POST /agent/chat ──► proxies to RAG backend │
│ SQLite (users.db) · Argon2id · JWT (python-jose) │
└──────────────────────────────┬──────────────────────────────┘
│ Internal HTTP
▼
┌─────────────────────────────────────────────────────────────┐
│ Adaptive RAG Backend (FastAPI :8000) │
│ │
│ POST /rag/query POST /rag/documents/upload │
│ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ LangGraph Workflow │ │
│ │ │ │
│ │ Query Classifier ─► [index] ─► FAISS Retriever │ │
│ │ │ │ │ │
│ │ │ ▼ │ │
│ │ │ Document Grader │ │
│ │ │ ↙ ↘ │ │
│ │ │ Relevant Irrelevant │ │
│ │ │ │ │ │ │
│ │ ├─► [general]│ ▼ │ │
│ │ │ Groq │ Query Rewriter │ │
│ │ │ Direct │ │ │ │
│ │ │ │ ▼ │ │
│ │ └─► [search]─┴──────► Web Search (Tavily) │ │
│ │ │ │ │
│ │ ▼ │ │
│ │ Generate Answer (Groq LLM) │ │
│ └─────────────────────────────────────────────────────┘ │
│ Gemini Embeddings (text-embedding-004) · FAISS · MongoDB │
└─────────────────────────────────────────────────────────────┘
| Feature | Details |
|---|---|
| 🔐 Authentication | JWT Bearer tokens · Argon2id password hashing · Register/Login |
| 🧠 Adaptive Routing | Classifier decides: RAG / General LLM / Web Search |
| 📄 Document Upload | PDF & TXT ingestion → chunking → Gemini embedding → FAISS |
| 🌐 Web Search Fallback | Tavily API for real-time information |
| 💬 Chat History | Per-user session persistence in MongoDB |
| ⚡ Groq-Powered | Llama 3.3 70B (Groq) for LLM inference · text-embedding-004 for embeddings |
| 🖥️ Streamlit UI | Glassmorphism dark theme with login, chat, and file upload |
Adaptive-Rag/
├── adaptive_rag/ # RAG backend + Streamlit frontend
│ ├── main.py # FastAPI entry point (:8000)
│ ├── requirements.txt
│ ├── src/
│ │ ├── api/routes.py # /rag/query, /rag/documents/upload
│ │ ├── config/
│ │ │ ├── settings.py # Environment config (Pydantic)
│ │ │ └── prompts.yaml # LLM prompt templates
│ │ ├── llms/groq_llm.py # Groq LLM client + Gemini embeddings
│ │ ├── memory/chat_history_mongo.py
│ │ ├── models/ # State, Grade, RouteIdentifier schemas
│ │ └── rag/
│ │ ├── graph_builder.py # Full LangGraph workflow
│ │ ├── retriever_setup.py # FAISS vector store
│ │ └── document_upload.py # File ingestion pipeline
│ └── streamlit_app/
│ ├── Home.py # Login / Register page
│ ├── pages/chat.py # Protected chat UI
│ └── utils/api_client.py # HTTP client (JWT Bearer)
│
├── auth_service/ # Authentication BFF (migrated from Rust)
│ ├── app/
│ │ ├── main.py # FastAPI entry point (:8080)
│ │ ├── auth.py # JWT + Argon2id
│ │ ├── database.py # Async SQLite (aiosqlite)
│ │ ├── models.py # SQLAlchemy User model
│ │ ├── schemas.py # Pydantic request/response schemas
│ │ ├── dependencies.py # JWT bearer dependency
│ │ └── routers/
│ │ ├── auth.py # POST /auth/register, /auth/login
│ │ ├── user.py # GET /user/me
│ │ └── agent.py # POST /agent/chat (proxy)
│ └── requirements.txt
│
├── .env.example # Environment variables template
├── .gitignore
└── README.md
Note: rename
src/llms/gemini.py→src/llms/groq_llm.py(or keep the filename and just swap internals) to match — update if you named it differently.
- Python 3.11+
- Groq API key (free) — for LLM inference
- Google AI Studio API key — for Gemini embeddings (
text-embedding-004) - Tavily API key (free tier available)
- MongoDB (optional — chat history is disabled gracefully if unavailable)
git clone https://github.com/itsmehotpants/Adaptive-Rag.git
cd Adaptive-Rag
cp .env.example .env
# Edit .env and fill in your GROQ_API_KEY, GEMINI_API_KEY (embeddings), and TAVILY_API_KEY
cd auth_service
python -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # macOS/Linux
pip install -r requirements.txt
uvicorn app.main:app --host 0.0.0.0 --port 8080 --reload
Auth service will be available at: http://localhost:8080 API docs: http://localhost:8080/docs
cd adaptive_rag
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --host 0.0.0.0 --port 8000 --reload
RAG backend at: http://localhost:8000 API docs: http://localhost:8000/docs
cd adaptive_rag/streamlit_app
streamlit run Home.py
Frontend at: http://localhost:8501
1. Register: POST /auth/register { username, password }
2. Login: POST /auth/login { username, password } → { access_token }
3. Use token: Authorization: Bearer <access_token>
4. Chat: POST /agent/chat { query } (proxied to RAG with session_id)
Note: Streamlit Community Cloud runs only the Streamlit app. You need the backend services deployed separately.
- Push your code to GitHub (already done).
- Go to share.streamlit.io.
- Click New app → Select
itsmehotpants/Adaptive-Rag. - Set Main file path:
adaptive_rag/streamlit_app/Home.py - Under Advanced → Secrets, add:
AUTH_SERVICE_URL = "https://your-auth-service.onrender.com"
RAG_SERVICE_URL = "https://your-rag-backend.onrender.com"
- Click Deploy.
- Go to render.com → New Web Service.
- Connect GitHub repo → Select
itsmehotpants/Adaptive-Rag. - Settings:
- Root Directory:
auth_service - Build Command:
pip install -r requirements.txt - Start Command:
uvicorn app.main:app --host 0.0.0.0 --port $PORT
- Add environment variables:
JWT_SECRET,ACCESS_TOKEN_EXPIRE_MINUTES=60
- New Web Service → same repo.
- Settings:
- Root Directory:
adaptive_rag - Build Command:
pip install -r requirements.txt - Start Command:
uvicorn main:app --host 0.0.0.0 --port $PORT
- Add environment variables:
GROQ_API_KEY,GEMINI_API_KEY,TAVILY_API_KEY,MONGODB_URL
# Install Railway CLI
npm install -g @railway/cli
railway login
# Deploy auth service
cd auth_service
railway up
# Deploy RAG backend
cd ../adaptive_rag
railway up
| Method | Endpoint | Auth | Description |
|---|---|---|---|
POST |
/auth/register |
❌ | Create new user account |
POST |
/auth/login |
❌ | Login and receive JWT token |
GET |
/user/me |
✅ JWT | Get current user profile |
POST |
/agent/chat |
✅ JWT | Proxy chat to RAG backend |
GET |
/health |
❌ | Health check |
| Method | Endpoint | Description |
|---|---|---|
POST |
/rag/query |
Run adaptive RAG query |
POST |
/rag/documents/upload |
Upload & index document |
GET |
/ |
Health check |
| Layer | Technology |
|---|---|
| LLM | Groq — llama-3.3-70b-versatile |
| Embeddings | Google text-embedding-004 |
| RAG Orchestration | LangGraph 0.5 |
| Vector Store | FAISS (in-memory) |
| Web Search | Tavily API |
| Chat History | MongoDB |
| Auth Backend | FastAPI + Argon2id + JWT |
| Auth Database | SQLite (aiosqlite) |
| RAG Backend | FastAPI + Uvicorn |
| Frontend | Streamlit |
MIT © 2024 itsmehotpants