Agent Based Reliable Augmented Generation for Medical Literature Summarization

Abstract

The exponential expansion of biomedical publications has created a persistent challenge of information overload for clinicians, researchers, and policy-makers. Manual review and synthesis of medical literature are increasingly impractical, while current automated summarization systems often suffer from hallucinations, limited factual grounding, and dependence on external cloud services that compromise data privacy and reproducibility. This paper presents an Agent-Based Reliable Retrieval-Augmented Generation (RAG) Framework designed to generate concise, evidence-grounded, and verifiable summaries of biomedical literature. The proposed system integrates multiple coordinated agents—Retriever, Summarizer, Fact-Checker, Citation Manager, and Reliability Evaluator—to ensure that each generated summary maintains factual accuracy and transparent citation linkage. Operating entirely in an offline environment, the framework preserves user privacy and supports reproducibility on standard academic hardware. Evaluation will employ benchmark biomedical datasets such as PubMed and BioASQ, with both lexical and faithfulness-oriented metrics, including ROUGE, BLEU, evidence-coverage ratio, hallucination rate, and citation accuracy. The framework aims to bridge the reliability gap between large language models and the stringent requirements of healthcare informatics, offering a trustworthy, reproducible, and ethically compliant solution for automated biomedical knowledge synthesis.