PolicyGuard: A RAG-based Insurance Claim Prediction and Policy Query System

Insurance policy documents often have complicated legal language, exclusions, and structures that are hard for people to understand. A lot of claims are turned down not because they aren’t eligible, but because they don’t understand them. This paper introduces PolicyGuard, a Retrieval-Augmented Generation (RAG) system that examines user-uploaded policy documents and forecasts the likelihood of acceptance or rejection of future claims. The framework integrates OCR, semantic chunking, embedding-based retrieval, and grounded LLM reasoning. Experimental results show improved transparency, clause-based citation, and reduced hallucination.