Abstract

This project implements a RAG-based (Retrieval-Augmented Generation)
AI assistant that answers questions based on Ready Tensor publications.
The system combines vector search with large language models to provide
accurate, context-aware responses grounded in real documents.

Overview

Traditional AI chatbots generate responses purely from training data,
which can lead to hallucinations or outdated information. This RAG-based
assistant solves that by first retrieving relevant content from a
document corpus, then using an LLM to generate answers based on that
retrieved context.

System Architecture

The pipeline works in the following steps:

1. Document Loading - Text files are loaded from the data directory
2. Text Chunking - Documents are split into 500-character chunks
3. Embedding - Chunks are converted to vectors using
   sentence-transformers (all-MiniLM-L6-v2)
4. Storage - Vectors are stored in ChromaDB
5. Retrieval - User queries are matched against stored vectors
6. Generation - Retrieved context + query is sent to an LLM
   for response generation

Tech Stack

• LLM: Groq (Llama 3.1)
• Vector Database: ChromaDB
• Embeddings: HuggingFace sentence-transformers
• Framework: LangChain
• Language: Python 3.12

Setup Instructions

1. Clone the repository
2. Install dependencies: