🤖 AI RAG Chatbot: Enterprise-Grade Knowledge Retrieval

A production-ready Retrieval-Augmented Generation (RAG) chatbot engineered to deliver precise, context-aware AI responses grounded in local document data.

Unlike basic "hello world" wrappers, this project implements a 3-stage advanced retrieval pipeline and features a microservices architecture fully containerized with Docker.

---

🎯 Why This Project Stands Out (For Engineering Teams)

I built this project to demonstrate a deep understanding of modern backend development and advanced LLM integration. Here is what sets it apart:

• Advanced Retrieval Pipeline: Moves beyond basic cosine similarity. It uses Hybrid Search (FAISS Vector + BM25 Keyword), Multi-Query Expansion (generating semantic variations of user queries), and FlashRank Re-ranking to achieve ~90% retrieval accuracy.
• Microservices Architecture: Clean separation of concerns. The FastAPI backend handles the heavy lifting, vector processing, and state management, while the Streamlit frontend acts purely as a stateless UI client.
• Production-Ready Containerization: Fully dockerized with docker-compose. Utilizes optimized multi-stage builds and the blazing-fast uv package manager to keep build times under 2 minutes. CPU-optimized PyTorch keeps the image sizes lean.
• Asynchronous & Typed: Written in Python 3.12 utilizing strict Pydantic typing and fully asynchronous endpoints (async/await) for high concurrency.
• Persistent Storage: Configured Docker volumes ensure that local vector stores and conversation memory survive container restarts.

🚀 Quick Start (Docker)

The absolute fastest way to run this application is via Docker. No virtual environments required.

Prerequisites

• Docker & Docker Compose installed.
• A free Groq API Key.
• (Optional) A Hugging Face Token to speed up embedding model downloads.

1. Configure Environment

Clone the repository and create a .env file in the root directory:

# Required
GROQ_API_KEY="your_groq_api_key_here"
GROQ_MODEL="llama-3.3-70b-versatile"

# Optional (Prevents HuggingFace rate-limit warnings)
HF_TOKEN="your_huggingface_token_here"

# Vector Store
VECTOR_STORE_PATH="./data/vectorstore"
EMBEDDING_MODEL="sentence-transformers/all-MiniLM-L6-v2"

2. Build and Run

Start the application using Docker Compose. We use uv inside the container, so the build is incredibly fast.

docker compose up --build -d

3. Access the Application

• Web UI (Streamlit): http://localhost:8501
• API Documentation (Swagger): http://localhost:8000/docs

To stop the application, run:

docker compose down

---

🧠 Architecture Overview

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   User Query    │────▶│  Multi-Query    │────▶│  Hybrid Search  │
│                 │     │  Expansion      │     │  (Vector+BM25)  │
└─────────────────┘     └─────────────────┘     └─────────────────┘
                                                           │
                                                           ▼
┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   Response to   │◀────│      LLM        │◀────│ FlashRank       │
│      User       │     │  (Groq/LLaMA 3) │     │  Re-ranking     │
└─────────────────┘     └─────────────────┘     └─────────────────┘
       │
       ▼
┌─────────────────┐
│ Conversation    │
│ Memory          │
└─────────────────┘

💻 Tech Stack

• Backend Framework: FastAPI, Uvicorn
• Frontend UI: Streamlit
• AI / LLM Orchestration: LangChain
• Inference API: Groq (Llama 3.3 70B)
• Embeddings & Re-ranking: Hugging Face (all-MiniLM-L6-v2), FlashRank
• Vector Database & Search: FAISS, Rank-BM25
• Environment Management & Building: uv, Docker, Docker Compose

🛠️ Local Development Setup (Without Docker)

If you prefer to run the application directly on your host machine for debugging or development:

# 1. Create a virtual environment using uv
uv venv

# 2. Activate it
# On Windows:
.venv\Scripts\activate
# On macOS/Linux:
source .venv/bin/activate

# 3. Install dependencies
uv pip install -e ".[dev]"

# 4. Run the backend (Terminal 1)
python run_api.py

# 5. Run the frontend (Terminal 2)
streamlit run ui/streamlit_app.py

🧪 Testing

The project uses pytest for unit and integration testing.

# Run all tests
pytest

# Run tests with coverage report
pytest --cov=src --cov-report=term-missing

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.