🤖 RAG Coding Assistant

A production-grade Retrieval Augmented Generation (RAG) app for coding documents.

🏗️ Architecture

frontend/ (Streamlit)
    ├── app.py              ← Login/Register gate + navigation
    ├── api_client.py       ← HTTP client for backend
    └── pages/
        ├── chat.py         ← Multi-session chat UI
        ├── documents.py    ← Upload & manage documents
        └── settings.py     ← User settings

backend/ (FastAPI)
    ├── main.py             ← App entry point
    ├── rag.py              ← Core RAG engine
    ├── config.py           ← All configuration
    ├── models.py           ← SQLAlchemy DB models
    ├── database.py         ← Async DB connection
    ├── auth.py             ← JWT authentication
    ├── schemas.py          ← Pydantic schemas
    └── routers/
        ├── auth.py         ← /api/auth/*
        ├── documents.py    ← /api/documents/*
        └── chat.py         ← /api/chat/*

⚡ Quick Start

1. Clone & Install

git clone <your-repo>
cd rag-project
pip install -r requirements.txt

2. Configure

cp .env.example .env
# Edit .env and add your GEMINI_API_KEY

Get free Gemini API key: https://aistudio.google.com

3. Run

Option A — Two terminals:

# Terminal 1: Backend
uvicorn backend.main:app --reload --port 8000

# Terminal 2: Frontend
streamlit run frontend/app.py

Option B — Single script:

chmod +x start.sh
./start.sh

Option C — Docker:

docker-compose up --build

4. Open

• 🌐 Frontend: http://localhost:8501
• 📖 API Docs: http://localhost:8000/docs

---

🔑 Features

Feature	Details
Auth	JWT login/register, per-user data isolation
File Support	PDF, TXT, MD, DOCX, CSV + URL scraping
Hybrid Search	Semantic (ChromaDB) + BM25 keyword search
Reranking	CrossEncoder for precision retrieval
Chat History	Multi-turn conversations with session memory
Multi-session	Create, switch, delete chat sessions
Source Viewer	See exactly which chunks were used
Streaming	Token-by-token response streaming endpoint
Docker	Full containerized deployment

---

🧠 RAG Pipeline

Upload PDF
  → PyMuPDF parse
  → Sentence-aware chunking (500 chars, 50 overlap)
  → SentenceTransformer embed (all-MiniLM-L6-v2)
  → ChromaDB store (per-user collection)

User Question
  → Embed question
  → ChromaDB semantic search (top 10)
  → BM25 keyword scoring
  → Combine scores (60% semantic + 40% BM25)
  → CrossEncoder rerank (top 5)
  → Build prompt with context + history
  → Gemini generate answer
  → Return answer + sources

---

🌐 API Endpoints

Method	Endpoint	Description
POST	/api/auth/register	Create account
POST	/api/auth/login	Login → token
GET	/api/auth/me	Current user
POST	/api/documents/upload	Upload file
POST	/api/documents/url	Scrape URL
GET	/api/documents/	List documents
DELETE	/api/documents/{id}	Delete document
GET	/api/chat/sessions	List sessions
POST	/api/chat/sessions	New session
POST	/api/chat/sessions/{id}/messages	Send message
GET	/api/chat/sessions/{id}/messages	Get history
POST	/api/chat/sessions/{id}/stream	Stream response

---

🚀 Deploy Free

Platform	Service	Command
Streamlit Cloud	Frontend	Connect GitHub repo
Railway.app	Backend	railway up
Render.com	Full stack	Deploy via dashboard
Hugging Face Spaces	Streamlit UI	Upload to Space