AI Policy RAG QA System

RAG system for semantic question answering over handbook-style policy documents, designed to make long documents easier to search, retrieve, and answer against.

Highlights

• Built document chunking, embedding, retrieval, and grounded QA end to end.
• Compared chunking strategies, similarity thresholds, and top-k retrieval settings.
• Evaluated multiple models to understand quality and cost tradeoffs.
• Focused on grounded answers rather than open-ended generation.

Tech Stack

Python, FAISS, sentence-transformers, PyMuPDF, LlamaIndex, and OpenAI API.

Why It Matters

This project shows the core design tradeoffs behind production RAG systems, especially around retrieval quality, chunking, and answer grounding.