GenRAG

GenRAG is a terminal tool designed to set up a Retrieval-Augmented Generation Pipeline locally from scratch, without utilizing any high-level frameworks like LangChain or vector databases. It includes features such as recursive text splitting, chunking, and building embeddings. The embeddings are stored in a CSV file,without using any Vector Databases and searching is based on cosine similarity.

Features

• Recursive text splitting
• Text chunking
• Building embeddings
• Storing embeddings in a CSV file
• Searching based on cosine similarity
• Inference from local LLM/API

Installation

1. Clone the repository:

   git clone https://github.com/adityaghai07/GenRAG.git
   cd GenRAG

2. Create and activate a virtual environment:

   python3 -m venv env
   source env/bin/activate  # On Windows use `env\Scripts\activate`

3. Install the required dependencies:

   pip install -r requirements.txt

Usage

1. Add a PDF file to the data folder and change the path in the script as necessary.

2. Create embeddings by running:

   python3 create_embeddings.py

   You should see a CSV file generated in the data folder.

3. Run the main script:

   python3 main.py

   Enter your query when prompted.

LLM Response

You can use both a local LLM or an LLM from an API like Gemini for generating responses.

• Local LLM: If you have the capability to run a local LLM, you can use it for generating responses. Cause mine is too slow :(

• LLM from API: If your system is not powerful enough for local inference, you can use an API like Gemini. To do this, create a .env file and pass the Gemini API key.

Using Gemini API

1. Create a .env file in the root directory of the project.
2. Add your Gemini API key to the .env file:

   GEMINI_API_KEY=your_api_key_here

3. The system will use the API key for generating responses through the Gemini API.

Credits

Special thanks to the following YouTube channels and research papers for their invaluable resources and insights:

YouTube Channels

• AI Anytime
• Daniel Brouke
• Krish Naik
• Codebasics
• CampusX

Research Papers

• Patrick Lewis ., "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks" arXiv:2005.11401
• Vaswani et al., "Attention is All You Need" arXiv:1706.03762
• Reimers et al., "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks" arXiv:1908.10084

License

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