🤖 Gromo FAQ RAG Chatbot with Web Search

A powerful Retrieval-Augmented Generation (RAG) chatbot for Gromo that provides intelligent responses by combining:

1. 📚 Company FAQ dataset - For accurate company-specific information
2. 🔎 Real-time web search results - Using SERP API for up-to-date information

✨ Features

• 🧠 Advanced LLM Support:

  • Mistral AI API (recommended for best performance)
  • Qwen QwQ model (local deployment)
  • Configurable for various models based on your needs

• 📊 Enhanced Context Understanding:

  • Implements RAG architecture using Langchain
  • Vector database for efficient similarity search
  • Contextual response generation with relevant information retrieval

• 🌐 Real-time Web Search Integration:

  • SERP API integration for current information
  • Blends FAQ knowledge with web search results
  • Provides citations and sources for information

• 🖥️ Multiple Interface Options:

  • Gradio UI with modern interface (recommended)
  • Streamlit UI alternative
  • FastAPI endpoints for application integration

• 🛠️ Developer-Friendly Design:

  • Modular code structure
  • Extensive documentation
  • Easy customization options

🗂️ Project Structure

gromo_rag/
├── data/                  # Directory for FAQ dataset
├── src/                   # Source code
│   ├── __init__.py
│   ├── config.py          # Configuration settings
│   ├── data_loader.py     # Load and process FAQ data
│   ├── embeddings.py      # Vector embeddings for RAG
│   ├── rag_chain.py       # RAG implementation with Langchain
│   ├── web_search.py      # Web search integration with SERP API
│   └── utils.py           # Utility functions
├── app_gradio.py          # Gradio application (recommended)
├── app.py                 # Streamlit application
├── app_fastapi.py         # FastAPI application
├── test_chatbot.py        # Script to test the chatbot
├── init_vector_store.py   # Script to initialize vector store
├── .env                   # Environment variables (not tracked by git)
├── requirements.txt       # Project dependencies
└── README.md              # Project documentation

🚀 Quick Start

1. Clone the repository

git clone https://github.com/yasboop/RAG-Websearch-Mistral.git
cd RAG-Websearch-Mistral

2. Install dependencies:

pip install -r requirements.txt

3. Set up environment variables:

Create a .env file with your API keys:

# Mistral AI API key
MISTRAL_API_KEY=your_mistral_api_key_here

# SERP API key for Google search
SERPAPI_API_KEY=your_serpapi_key_here

4. Initialize the vector store:

python init_vector_store.py

5. Run the application:

# Using Gradio (recommended)
python app_gradio.py

# Using Streamlit
streamlit run app.py

# Using FastAPI
python app_fastapi.py

🧪 Testing

Test the chatbot functionality with sample questions:

python test_chatbot.py

Or test with a specific question:

python test_chatbot.py "What financial products does Gromo offer?"

⚙️ Customization

Customize the chatbot by modifying the configuration in src/config.py:

• Model Settings:

  • Use Mistral AI API: USE_MISTRAL_API = True
  • Use local Qwen model: USE_MISTRAL_API = False and USE_QUANTIZED_MODEL = False
  • Use quantized model for AWS deployment: USE_QUANTIZED_MODEL = True

• RAG Parameters:

  • Adjust context window size
  • Modify similarity search parameters
  • Fine-tune retrieval strategy

• Web Search Options:

  • Enable/disable web search
  • Change search result count
  • Adjust search relevance parameters

🔌 API Usage

The chatbot can be integrated into other applications through the FastAPI interface:

1. Start the API server:

   python app_fastapi.py

2. Send requests to the API:

   curl -X POST http://localhost:8000/chat \
     -H "Content-Type: application/json" \
     -d '{"query": "What is Gromo?"}'

🔑 Getting a SERP API Key

To use the web search functionality:

1. Visit https://serpapi.com/ and sign up for an account
2. Navigate to your dashboard to find your API key
3. Add the API key to your .env file as SERPAPI_API_KEY

🌟 Advanced Usage

AWS Deployment

This project includes scripts for AWS EC2 deployment. See AWS_DEPLOYMENT.md for detailed instructions.

Custom Knowledge Base

To use a different knowledge base:

1. Replace the CSV file in the data/ directory
2. Update the data loading logic in src/data_loader.py if necessary
3. Reinitialize the vector store with python init_vector_store.py

📄 License

MIT