Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{195855,
        author = {Mukesh Gilda and Gollena Adith and Nakka Jashwanth and Mahesh Bhati and Raj Goel},
        title = {MedGPT: An Agentic Medical AI Assistant with Hybrid Retrieval, Chain-of-Thought Reasoning, and NLI-Based Citation Mapping for Medical Education},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {1187-1195},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=195855},
        abstract = {MedGPT is an advanced agentic medical AI assistant designed to provide evidence-based clinical answers with transparent reasoning and verifiable citations. The system integrates a multi-stage retrieval-augmented generation (RAG) pipeline combining query expansion, hybrid retrieval (dense vector search + BM25 keyword matching + real-time PubMed), and cross-encoder reranking. A novel Natural Language Inference (NLI) citation mapper automatically links each answer sentence to supporting source documents and attaches sentence-level confidence scores. The knowledge base is grounded in Harrison’s Principles of Internal Medicine (approximately 18,000 indexed chunks) and augmented with live PubMed research. The reasoning backbone employs Llama 3.1 70B Instruct to produce structured six-step chain-of-thought (CoT) traces. Evaluation on a USMLE-style benchmark yields 70% accuracy; retrieval precision reaches 92%; citation accuracy is 85%. A controlled user study with 30 medical students records a statistically significant 16% improvement in post-test scores (p < 0.05) and an 82.3/100 System Usability Scale rating. The system promotes higher-order thinking skills (HOTS) aligned with Bloom’s Taxonomy Levels 4–6 and operationalises heutagogical self-directed learning. MedGPT addresses critical gaps in existing medical AI tools by providing transparent, multi-source, evidence-grounded assistance to medical students and healthcare professionals.},
        keywords = {Retrieval-Augmented Generation; Chain-of-Thought Reasoning; Natural Language Inference; Medical Education; USMLE Benchmark; Higher-Order Thinking Skills; Heutagogy; Large Language Models.},
        month = {April},
        }