Retrieval Augmented Generation from Scratch: Inception RAG

This repository contains the code for a simple Retrieval Augmented Generation (RAG) system built from scratch using Python and open-source tools. It uses Llama 3 as a language model, powered by Groq, and Nomic embeddings. It includes Christopher Nolan's Inception movie script as a source document in the data/docs folder to build the RAG app, but you can add any documents of your choice.

For a detailed explanation of the code and the concepts behind RAG, check out this blog post.

This project was developed by CodeAwake.

Structure

The code is contained in the app/ folder and organized into the following files:

• config.py: Configuration settings for the application.
• loader.py: Loads and processes the pdf source documents for RAG (chunking, embedding and storing in vector store).
• splitter.py: Text chunking functionality.
• vector_store.py: Simple vector store implementation.
• rag.py: Core RAG functionality and interactive Q&A loop.

Installation

Prerequisites ✅

• Python 3.11 or higher
• Poetry (Python package manager)

Instructions

1. Install the dependencies using Poetry:

   poetry install

2. Create a .env file copying the .env.example file provided and set the required environment variable:

   • GROQ_API_KEY: Your Groq API key.

Running the Application

To load the source pdf documents and process them for RAG:

poetry run load-docs

To run the interactive Q&A RAG app:

poetry run rag