Capstone - RAG System

Retrieval-Augmented Generation system for intelligent document Q & A. We're implementing 3 different approaches using a shared backend interface. React + TypeScript for frontend, Python at the backend for retrieval, orchestration, and LLM-based answer generation.

Project Goal

Build a document Q&A system that:

• supports document upload and querying,
• compares multiple retrieval strategies fairly,
• returns grounded answers with citations,
• and provides an evaluation area for benchmarking retrieval quality.

Core Components

• frontend/ — React + TypeScript user interface
• backend/ — Python backend and integrated RAG pipeline
• vector_retrieval/ — vector retrieval research and implementation
• keyword_retrieval/ — keyword/BM25 retrieval research & implementation
• clara_retrieval/ — CLaRa retrieval research and implementation
• shared_data/ — shared schemas, prompts, evaluation assets, API contracts, and docs

High-Level Flow

User → Frontend → Backend API → Retrieval Method → Top-k Chunks → LLM Generation → Answer + Citations

Tech Stack

Frontend

• React
• TypeScript
• Vite
• Axios
• Tailwind / Material UI

Backend

• Python
• FastAPI

RAG Stack

• SentenceTransformers
• FAISS or ChromaDB
• BM25
• OpenAI API or another LLM

Notes

This repository is structured to support both independent team development and final system integration.