RAG AI Agent with Pydantic AI and Supabase

A simple Retrieval-Augmented Generation (RAG) AI agent using Pydantic AI and Supabase with pgvector for document storage and retrieval.

Features

• Document ingestion pipeline for TXT and PDF files
• Vector embeddings using OpenAI
• Document storage in Supabase with pgvector
• Pydantic AI agent with knowledge base search capabilities
• Streamlit UI for document uploads and agent interaction

Project Structure

foundational-rag-agent/
├── database/
│   └── setup.py          # Database setup and connection utilities
├── document_processing/
│   ├── __init__.py
│   ├── chunker.py        # Text chunking functionality
│   ├── embeddings.py     # Embeddings generation with OpenAI
│   ├── ingestion.py      # Document ingestion pipeline
│   └── processors.py     # TXT and PDF processing
├── agent/
│   ├── __init__.py
│   ├── agent.py          # Main agent definition
│   ├── prompts.py        # System prompts
│   └── tools.py          # Knowledge base search tool
├── ui/
│   └── app.py            # Streamlit application
├── tests/
│   ├── test_chunker.py
│   ├── test_embeddings.py
│   ├── test_ingestion.py
│   ├── test_processors.py
│   └── test_agent.py
├── .env.example          # Example environment variables
├── requirements.txt      # Project dependencies
├── PLANNING.md           # Project planning document
├── TASK.md               # Task tracking
└── README.md             # Project documentation

Setup

1. Clone the repository
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. Copy .env.example to .env and fill in your API keys and configuration
5. Run the Streamlit application:

   streamlit run ui/app.py
   

6. Run the SQL in rag-example.sql to create the table and matching function for RAG

Usage

1. Upload documents (TXT or PDF) through the Streamlit UI
2. Ask questions to the AI agent
3. View responses with source attribution

Deployment with Docker on Render

To deploy this application on Render, build the provided Dockerfile and set th e service command to:

streamlit run ui/app.py --server.port $PORT --server.address 0.0.0.0

Render automatically injects the PORT environment variable which the container uses to expose the Streamlit UI.

Dependencies

• Python 3.11+
• Pydantic AI
• Supabase
• OpenAI
• PyPDF2
• Streamlit

License

MIT