Agentic RAG Chat

A fully custom chatbot built with Agentic RAG (Retrieval-Augmented Generation), combining Gemini models (free tier) with a local knowledge base for accurate, context-aware, and explainable responses. Features a lightweight, dependency-free frontend and a streamlined FastAPI backend for complete control and simplicity.

Features

• Pure HTML/CSS/JavaScript frontend with no external dependencies
• FastAPI backend with Gemini integration
• Agentic RAG implementation with:
  • Context retrieval using embeddings and cosine similarity
  • Step-by-step reasoning with Chain of Thought
  • Function calling for dynamic context retrieval
• Comprehensive error handling and logging
• Type-safe implementation with Python type hints
• Configurable through environment variables

Project Structure

agentic_rag/
├── backend/
│   ├── embeddings.py    # Embedding and similarity functions
│   ├── rag_engine.py    # Core RAG implementation
│   └── server.py        # FastAPI server
├── frontend/
│   └── index.html       # Web interface
├── requirements.txt     # Python dependencies
├── .env.sample         # Sample environment variables
└── README.md           # Documentation

Prerequisites

• Python 3.11 or higher
• Gemini API key (you can get one for free at google ai studio)
• Git (for version control)

Installation

1. Clone the repository:

git clone https://github.com/AndrewNgo-ini/agentic_rag.git
cd agentic_rag

2. Create and activate a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Python dependencies:

pip install -r requirements.txt

4. Set up environment variables:

cp .env.sample .env

Then edit .env with your configuration:

GEMINI_API_KEY=your-api-key-here
GEMINI_MODEL=gemini-2.0-flash  # or another compatible model
GEMINI_EMBEDDING_MODEL=text-embedding-3-small # or another compatible model

Running the Application

1. Start the backend server:

cd backend
python server.py

2. Access the frontend:

• Option 1: Open frontend/index.html directly in your web browser
• Option 2: Serve using Python's built-in server:

cd frontend
python -m http.server 3000

Then visit http://localhost:3000 in your browser.

Usage

1. Type your question in the input field
2. The system will:
   • Retrieve relevant context using embeddings
   • Generate a step-by-step reasoning process
   • Provide a final answer based on the context and reasoning
3. View the intermediate steps and reasoning process in the response

Configuration

The application can be configured through environment variables:

Variable	Description	Default
GEMINI_API_KEY	Your Gemini API key	Required
GEMINI_MODEL	Gemini model to use	gemini-2.0-flash
HOST	Backend server host	0.0.0.0
PORT	Backend server port	8000

Error Handling

The application includes comprehensive error handling:

• API errors are logged and returned with appropriate status codes
• Frontend displays user-friendly error messages
• Detailed logging for debugging and monitoring
• Graceful fallbacks for common failure scenarios

License

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

Troubleshooting

Common issues and solutions:

1. Gemini API Error

   • Verify your API key is correct
   • Check your API usage limits
   • Ensure the model name is valid

2. Backend Connection Failed

   • Confirm the backend server is running
   • Check the port is not in use
   • Verify firewall settings

3. Embedding Errors

   • Ensure input text is not empty
   • Check for proper text encoding
   • Verify numpy installation

Acknowledgments

• Gemini for their API and models
• FastAPI framework
• Contributors and maintainers