Transformer Paper RAG

A local Retrieval Augmented Generation (RAG) system that lets you ask questions about the famous research paper "Attention Is All You Need" using Groq LLM, FAISS, and Sentence Transformers.

Overview

This project builds an intelligent questio answering system over the Transformer paper.

Instead of sending the full PDF directly to an LLM, the system:

• Extracts text from the PDF
• Splits it into smaller chunks
• Converts each chunk into vector embeddings
• Stores embeddings in a local FAISS vector database
• Retrieves the most relevant chunks based on user queries
• Sends retrieved context + question to Groq LLM for accurate answers

This improves answer quality, reduces token usage, and keeps responses grounded in the source document.

Features

• PDF-based question answering
• Local vector search using FAISS
• Semantic embeddings using Sentence Transformers
• Fast inference with Groq LLM
• CLI-based interactive chat system
• Persistent vector storage

Tech Stack

• Python
• Groq API
• FAISS
• Sentence Transformers
• PyPDF
• NumPy
• Pickle

Project Structure

transformer-paper-rag/

├── data/

│ └── attention_is_all_you_need.pdf

├── store/

│ ├── faiss_index

│ └── chunks

├── init.py

├── ingest.py

├── retriever.py

├── rag_chain.py

├── agents.py

├── main.py

└── README.md

Install dependencies:

---bash pip install -r requirements.txt

Create a .env file:

---env GROQ_API_KEY=your_api_key_here

Usage

Step 1: Ingest the PDF

---bash python app\ingest.py

This will:

• Read the PDF
• Create text chunks
• Generate embeddings
• Store vectors in FAISS

Step 2: Start Chat

---bash python app\main.py

Now ask questions like:

• What is self attention?
• Why do transformers remove recurrence?
• What is positional encoding?

Use Cases

• Research paper understanding
• Academic learning
• Local document Q&A
• RAG pipeline learning