Generative AI SQL Chatbot

Repository

Repository: geoffgin/GenAI-SQL-Chatbot

Features

• Multiple interaction methods (RAG, TAG)
• LLM provider selection (OpenAI, Claude)
• Intent classification (Details in Classifier README)
• Vector search with PGVector
• Langfuse Analytics
• Conversation memory (until browser refresh)
• Docker-based deployment

Prerequisites

• Docker and Docker Compose
• Python 3.9+
• OpenAI API key
• Anthropic API key
• Langfuse account (optional)

Installation

1. Clone the repository and navigate to the directory:

   git clone https://github.com/geoffgin/GenAI-SQL-Chatbot.git
   cd GenAI-SQL-Chatbot

2. Configure environment variables: Copy .env.example to .env and fill in your API keys and configurations.
3. Build and start the Docker services (one-off):

   make run

4. After the installation, simply run:

   make up

5. Or run the application in developer mode:

   make dev

6. Shut down the application:

   make down

Chatbot Usage

Go to http://localhost:8501 for the main chatbot interface.

• Select your preferred interaction method (RAG, TAG)
• Choose an LLM provider (OpenAI or Claude)
• Start asking questions about your database

Go to http://localhost:3000 for the Langfuse interface when not running on dev mode.

Architecture

• Frontend: Streamlit
• Document Parsing: Docling
• Vector Database: PostgreSQL with pgvector
• Observability: Langfuse
• LLM Framework: LlamaIndex
• Container Orchestration: Docker Compose

Paper References

• RAG (Retrieval-Augmented Generation): Paper by Facebook AI
• TAG (Table-Augmented Generation): Paper by UC Berkeley & Stanford University

Data Source Statement

This project uses the Chinook database, a media store database, for development and testing purposes. However, it can be easily adapted for any enterprise or domain-specific use case.

• Chinook Database:
  • Ownership: Maintained by lerocha
  • Licenses and Use: The Chinook Database allows use, distribution, and modification without any warranty of any kind.
  • Access: Available on GitHub at Chinook Database
• Intent Classifier Data:
  • GretelAI Synthetic Text-to-SQL:
    • Ownership: Gretel.ai
    • Licenses and Use: Licensed under the Apache License 2.0, permitting use, distribution, and modification with proper attribution.
    • Access: Available on Hugging Face at GretelAI Synthetic Text-to-SQL
  • Factoid WebQuestions Dataset:
    • Ownership: WebQuestions (Berant et al., 2013, CC-BY)
    • Licenses and Use: Distributed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license, allowing sharing and adaptation with appropriate credit.
    • Access: Available on GitHub at Factoid WebQuestions Dataset

Evaluation Framework

Located under the eval sub-folder.