🤖 GENAI-RAG-BOT

A production-ready Generative AI Assistant built with Retrieval-Augmented Generation (RAG) architecture using FastAPI, ChromaDB, and LLMs (e.g., Mistral). This bot answers domain-specific queries using a vector store and never hallucinates — it only replies using relevant documents from its database.

---

🚀 Features

• 🔍 Context-aware Retrieval via ChromaDB
• 📚 LangChain Integration: Seamlessly chains LLM calls with retrieval, enabling advanced workflows like summarization, Q&A, and document parsing.
• 📚 Document Embeddings: Store and retrieve documents in vector format using cohere (embed-english-v3.0)
• 🧠 LLM-Powered Responses (e.g., Mistral via mistral.ai)
• 🔐 Strict: Only responds using data retrieved from vector store
• 💾 Source-aware: Supports citation-style responses
• 🧼 Clean FastAPI backend with CORS support
• ⚙️ Ready for deployment on Azure App Service
• 📦 Modular Router + Configured Error Handlers + Health Endpoint

---

🧠 How It Works

1. User query → /api/chat
2. Query embedding → Compare with ChromaDB document embeddings
3. Retrieve top-k chunks → Pass to the LLM as context
4. Generate response → Only if relevant chunks exist
5. No relevant data? → “Sorry, I don’t have any information about this.”

---

⚙️ Environment Variables

Create a .env file:

PORT=your-port
HOST=http://localhost:5000
CORS_URL=http://localhost:5173
JWT_SECRET=your-secret
MONGO_URL=your-mongo-url
MISTRAL_MODEL_NAME=your-model-name
MISTRAL_API_KEY=your-mistral-api-key
COHERE_API_KEY=your-cohere-api-key

---

💻 Local Development

# 1. Clone the repo
git clone https://github.com/amankrs21/GENAI-RAG-BOT.git
cd GENAI-RAG-BOT

# 2. Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Run the server
uvicorn main:app --reload

Then visit: http://localhost:5000/health

---

🛡️ Agent Behavior

This bot does not hallucinate.

It only replies based on retrieved chunks from the ChromaDB vector store. If nothing relevant is found, it replies: “Sorry, I don’t have any information about this.”

---

📦 Requirements

• Python 3.9+
• FastAPI
• Uvicorn
• ChromaDB
• Requests
• Uplinks for Mistral/LLM calls

(See requirements.txt for full list)

---

📄 License

This project is licensed under the Apache 2.0 License. See the LICENSE file for details.

---

🙋‍♂️ Author

Built with ❤️ by @amankrs21

---

✨ Want to Contribute?

PRs are welcome! This is just the beginning of RAG-native GenAI tooling.