Hybrid Search With Postgres & PgVector For RAG using Groq

This project demonstrates how to implement a hybrid search engine for Retrieval-Augmented Generation (RAG) using Postgres with PgVector. It showcases the use of asynchronous streaming with Groq's function calling capabilities in a FastAPI application.

Description

This project utilizes Postgres as a vector database to create a hybrid search engine that combines vector search and text search capabilities. It leverages PgVector for storing and querying vector embeddings, and Groq's Large Language Model for function calling to retrieve information from the database through hybrid search.

Technologies Used

Features

• FastAPI-based implementation
• Hybrid search combining vector and text search capabilities
• Function calling for executing commands and retrieving information
• Chat interface for real-time communication
• Product search and recommendation system

Getting Started

Prerequisites

• Python 3.8 or higher
• PostgreSQL
• PgVector

Dependencies

Refer to requirements.txt for a complete list of dependencies. Key dependencies include:

• FastAPI
• OpenAI API (for generating embeddings)
• Groq (for Large Language Model)
• SQLAlchemy
• PgVector

Installation

1. Clone the repository

2. Create and activate a virtual environment

3. Install dependencies:

   pip install -r requirements.txt
   

4. Install PostgreSQL:

   • For Ubuntu: sudo apt-get install postgresql
   • For macOS: brew install postgresql
   • For Windows: Download and install from the official PostgreSQL website

5. Create a database named rag_example_schema:

   psql -U postgres
   CREATE DATABASE rag_example_schema;
   

6. Enable the vector extension:

   \c rag_example_schema
   CREATE EXTENSION IF NOT EXISTS vector;
   \q
   

7. Set up environment variables: Create a .env file in the root directory and add the following variables:

   OPENAI_API_KEY=your_openai_api_key
   GROQ_API_KEY=your_groq_api_key
   DATABASE_NAME=rag_example_schema
   DATABASE_USER=your_database_user
   DATABASE_PASSWORD=your_database_password
   DATABASE_URL=localhost
   DATABASE_PORT=5432
   

8. Temporarily disable the index in models/product.py: Comment out the following lines:

   # index_ada002 = Index(
   #     "hnsw_index_for_innerproduct_product_embedding_ada002",
   #     Product.embedding,
   #     postgresql_using="hnsw",
   #     postgresql_with={"m": 16, "ef_construction": 64},
   #     postgresql_ops={"embedding_ada002": "vector_ip_ops"},
   # )

9. Load initial data:

   python scripts/load_data.py
   

10. Re-enable the index in models/product.py: Uncomment the lines you commented in step 8:

    index_ada002 = Index(
        "hnsw_index_for_innerproduct_product_embedding_ada002",
        Product.embedding,
        postgresql_using="hnsw",
        postgresql_with={"m": 16, "ef_construction": 64},
        postgresql_ops={"embedding_ada002": "vector_ip_ops"},
    )

11. Run the application:

    uvicorn main:app --reload
    

Your application should now be running at http://localhost:8000.

Contributing

Contributions are welcome! Please open an issue or submit a pull request with your changes.