This research proposes a Retrieval-Augmented Generation (RAG)-based multi-module AI system designed to streamline interaction with health insurance information. Unlike prior approaches that treat conversational assistance, policy recommendation, and document retrieval as isolated tasks, our system unifies these modules into a single architecture. The framework integrates a chatbot for general insurance queries, a policy recommendation engine leveraging RAG with both structured and unstructured policy data, and a document retrieval module for clause-level search from uploaded policies. A distinct contribution is the inclusion of an evaluator agent that simulates human judgment to assess response quality across relevance, accuracy, clarity, and helpfulness-providing an automated feedback loop to improve performance over time. Experimental results demonstrate strong semantic retrieval (BERTScore F1 up to 0.84), robust recommendation capability (Hit@5 = 1.0, Recall@5 = 0.833), and effective clause retrieval from policy documents (BERTScore F1 = 0.8443). The novelty of this work lies in the domain-specific application of RAG with a modular architecture and quality-assessment agent, offering reduced hallucination risk, improved policy transparency, and user-focused insurance support.
Hanmante et al. (2025) studied this question.