Chainlit App with Query Decomposition and Hybrid Search

This project is a Chainlit application built using LangChain, incorporating features such as query decomposition and hybrid search. It enables users to upload PDF documents, break down complex queries into simpler sub-questions, and fetch answers using a combination of semantic search and BM25 retrieval techniques.

Features

• Query Decomposition: Automatically breaks down complex queries into simpler sub-questions.
• Hybrid Search: Combines semantic search using vector embeddings (Chroma) and keyword search (BM25) for efficient document retrieval.
• Azure Document Intelligence: Processes PDFs using Azure AI Document Intelligence to extract meaningful chunks for better searchability.
• Conversational Interface: Built on Chainlit for an interactive chat-based user experience.
• Memory-Powered Conversations: Utilizes LangChain's memory capabilities to maintain chat history and improve conversational context over time.

Project Structure

📁 your-repo/ │ ├── 📄 .gitignore # Specifies untracked files to be ignored by Git ├── 📄 README.md # This README file ├── 📄 pyproject.toml # Poetry configuration file for managing dependencies ├── 📄 chainlit_config.py # Chainlit app configuration file └── 📁 src/ # Source code directory ├── 📄 main.py # Main application logic └── 📁 assets/ # Assets like PDF files and logs (if applicable)

Prerequisites

Before setting up the project, make sure you have the following installed:

• Python 3.12.x (The project is tested on Python 3.11.9)
• Poetry (for dependency management)

Install Poetry

If you don't have Poetry installed, you can install it via the command:

curl -sSL https://install.python-poetry.org | python3 -

Setup

1. Clone the Repository

 git clone https://github.com/your-username/your-repo.git
  cd your-repo

2. Install Dependencies

Use Poetry to install the project dependencies:

poetry install

This command reads the pyproject.toml file to install the necessary dependencies, including LangChain, Chainlit, and Azure integrations.

3. Set up Environment Variables

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

AZURE_OPENAI_API_KEY=your_openai_api_key
AZURE_OPENAI_API_VERSION=2023-03-15-preview
AZURE_OPENAI_CHAT_DEPLOYMENT_NAME=your_deployment_name
AZURE_OPENAI_ENDPOINT=https://your-openai-endpoint.azure.com
AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT=https://your-document-intelligence-endpoint.azure.com
AZURE_DOCUMENT_INTELLIGENCE_KEY=your_document_intelligence_key

4. Run the App

Start the Chainlit app with the following command:

poetry run chainlit run src/4_qa_qery_decomp_hybrid.py

5. Upload PDFs and Ask Questions

• After starting the app, you will be prompted to upload PDF files.
• Once the files are processed, you can ask complex questions.
• The app will break the question into simpler sub-questions, search for answers in the documents using both BM25 and semantic search, and provide comprehensive answers with source references.