Advanced Rag with LangRaph and RAG tool

This project is an AI-powered assistant for question-answering tasks. It uses RAG tool and models to retrieve and process information from a PDF document.

Table of Contents

• Difference Between Advanced RAG and Normal RAG
• AI Question Refactoring and Subdivision, with example
• Installation
• Usage
• Project Structure
• Contributing
• License

Difference Between Advanced RAG and Normal RAG

Normal RAG (Retrieval-Augmented Generation):

• Combines retrieval and generation models.
• Retrieves relevant documents from a knowledge base.
• Generates answers based on the retrieved documents.
• Typically uses a single retrieval step followed by a generation step.

Advanced RAG:

• Enhances the basic RAG framework with additional features.
• May include multiple retrieval steps to refine the search results.
• Incorporates conditional logic to handle complex queries.
• Utilizes advanced techniques like re-ranking and filtering to improve answer quality.
• Can integrate with external tools and APIs for more comprehensive information retrieval.

AI Question Refactoring and Subdivision

In this project, the AI can refactor and subdivide complex questions into simpler sub-questions. It then calls the RAG tools with these sub-questions and appropriate prompts to provide a comprehensive answer.

Example

If you prompt the AI with:

User: Tell me about Lakshya's experience, projects, and education.

The AI will refactor this into three sub-questions:

1. What is Lakshya's experience?
2. What are Lakshya's projects?
3. What is Lakshya's education?

The AI then calls the RAG tools with these sub-questions and combines the responses to provide an elaborate answer:

Lakshya has nearly 6 years of experience in the field of blockchain, cloud, and full-stack solutions. His roles have included working as a full-stack developer, a blockchain and DevOps engineer, and most recently as a web3 solution architect and team lead for a fund tokenization project. During his tenure, he has worked with numerous clients and technologies, including NodeJS, ReactJS, Hyperledger fabric, and Hyperledger Besu.

Regarding projects, Lakshya has worked on a number of them, including:
1. Hyperledger/bevel - An accelerator that aids developers in setting up and deploying a production-ready DLT network.
2. Turbulent - A React.js and Next.js hook library for blockchain interface, available on NPM.
3. Enthalpy - A react UI kit available on NPM.
4. Taming the Ethereum Jungle: Wallact Makes Smart Contracts Your Plaything - A project involving Ethereum smart contracts.
5. Worked on creating Ripple APIs for banking clients.

As for his education, he holds a Bachelor of Technology (BTech) degree from Delhi Technological University (DTU, Delhi) obtained in 2015. He also has a Master of Technology (MTech) degree in Computational Mathematics from the National Institute of Technology (NIT), Surathkal, Karnataka which he earned between 2016 and 2018.

Installation

1. Clone the repository:

   git clone https://github.com/lakshyakumar/advanced-rag.git
   cd your-repo

2. Create and activate a virtual environment:

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

3. Install the required dependencies:

   pip install -r requirements.txt

4. Create a .env file based on the env.sample file and fill in the necessary environment variables.

Usage

1. Run the main script:

   python main.py

2. Interact with the assistant by typing your questions. To exit, type quit, exit, or q.

Project Structure

• main.py: The main script that sets up and runs the assistant.
• assets/: Directory containing the PDF document to be processed.
• requirements.txt: List of dependencies required for the project.
• venv/: Virtual environment directory.

Contributing

Contributions are welcome! Please open an issue or submit a pull request for any changes.

License

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