Single-Node RAG Integration with Llama 3 for European AI Act PDF Q&A

Description

We aim to integrate a single-node Retrieval-Augmented Generation (RAG) system with Llama 3 to facilitate Q&A capabilities for a combination of various sources of the European AI Act documents. The objective is to enable efficient and accurate question-answering from these documents.

Additionally, we seek to evaluate the performance and understand the answers provided by various open-source vector stores and embeddings for the different PDF files involved.

Interesting Library - RAGxplorer

https://github.com/gabrielchua/RAGxplorer?tab=readme-ov-file
pip install ragxplorer

Tasks

Done

• Built a simple single-node RAG-System
• Answers from FAISS and Chrome

To-do

• Connect it with LLM-hosted Llama-3
• Containerize and enhance UI

Improvizations

• Enhance RAG system training with additional PDF documents
• Dynamic chunking of PDF data to ensure information consistency
• Benchmark the performance of FAISS and Chrome
• Draw inspiration from RAGxplorer