Volume 2 - Issue 3, May - June 2026

The rapid expansion of digital health information and the complexity of clinical decision-making have generated a pressing demand for intelligent, evidence-grounded healthcare information systems. Conventional Large Language Models (LLMs) exhibit impressive generative fluency but remain vulnerable to hallucination, static knowledge boundaries, and inability to attribute responses to verifiable sources — characteristics rendering them unsuitable for unsupervised deployment in safety-critical clinical environments. This paper introduces MedRAG, a healthcare conversational assistant embedding a Retrieval-Augmented Generation (RAG) pipeline at its core. MedRAG dynamically retrieves current evidence-based content from curated medical knowledge repositories before generating any clinical response, ensuring every output is grounded in authoritative literature rather than static parametric memory. The system is designed and evaluated across three architectural paradigms of progressive sophistication: Foundational RAG, Optimised RAG, and Modular RAG. Experiments conducted over five standardised medical benchmarks — MedQA-USMLE, MedMCQA, PubMedQA, MMLUMedical, and a private chronic-pain diagnostic dataset — demonstrate consistent and statistically significant improvements over standalone LLM baselines. The Modular RAG configuration with GPT-4-Turbo achieves 91.1% accuracy on MedQA-USMLE and nearzero hallucination rates measured by the RAGAS faithfulness metric. Open-source models augmented with the modular pipeline record accuracy gains exceeding 27 percentage points, substantially narrowing the performance gap to proprietary frontier systems at a fraction of computational cost.

Alur Taher Basha, Mr.P.Bharath Kumar, Dr.D.William Albert,"MedRAG: A Healthcare Conversational Assistant Using Retrieval-Augmented Generation" International Journal of Advanced Multidisciplinary Research and Educational Development, V2(3): Page(709-714) May-June 2026. ISSN: 3107-6513. www.ijamred.com. Published by Scientific and Academic Research Publishing.