PRODUCTION-READY RETRIEVAL-AUGMENTED GENERATION FRAMEWORK FOR HEALTHCARE CLAIMS DECISION SUPPORT

Abstract

Healthcare claims processing involves complex clinical documentation, medical coding standards, payer policies, and regulatory compliance requirements. Conventional decision support systems primarily rely on predefined business rules and keyword-based information retrieval, often limiting their ability to provide accurate, context-aware recommendations for claims adjudication and prior authorization. Retrieval-Augmented Generation (RAG) has emerged as an effective artificial intelligence paradigm that combines large language models with external enterprise knowledge repositories to generate trustworthy, evidence-based responses. This paper proposes a ProductionReady Retrieval-Augmented Generation framework that integrates enterprise document indexing, semantic search, vector databases, knowledge grounding, automated model governance, and MLOps pipelines for healthcare claims decision support. The proposed framework enables intelligent retrieval of payer policies, coding guidelines, clinical documentation, and historical claims while minimizing hallucinations through retrievalbased evidence generation. Experimental evaluation demonstrates improvements in retrieval accuracy, decision consistency, response latency, operational efficiency, and regulatory compliance. The proposed architecture provides a scalable and production-ready solution for intelligent healthcare claims management. Keywords— Retrieval-Augmented Generation, Healthcare Claims, Decision Support, Large Language Models, Vector Database, Semantic Search, MLOps, Healthcare AI.