SHM Converge Scientific Abstract ePoster Gallery-Vlrchat: A Large Language Model Chatbot That Leverages Retrieval-Augmented Generation (Rag) to Answer Clinical Questions

Vlrchat: A Large Language Model Chatbot That Leverages Retrieval-Augmented Generation (Rag) to Answer Clinical Questions

Back to course

Pdf Summary

VLRChat is an AI-powered educational decision support tool designed to answer clinical questions using the Vanderbilt Internal Medicine (VIM) Handbook as its knowledge base. The project addresses a common problem in patient care: clinicians and trainees often search multiple sources for answers, yet many questions remain unanswered. By using retrieval-augmented generation (RAG), VLRChat provides real-time, evidence-based responses while reducing hallucinations and improving trustworthiness.<br /><br />Unlike generic chatbots, VLRChat is hyperlocal. It can deliver Vanderbilt-specific guidance on clinical management and institutional policies, making it more useful for local practice. The system also tracks user queries, allowing the team to identify knowledge gaps and opportunities for improvement in medical education content.<br /><br />In a study using 180 clinical questions generated by four users, VLRChat successfully answered 150 questions, for a success rate of 83%. Human evaluators rated its performance highly across several domains on a 5-point scale: accuracy 4.5, relevance 4.7, completeness 4.6, and clarity 4.8.<br /><br />Overall, VLRChat appears to effectively connect clinicians with reliable, concise information and improve educational decision support. The authors plan to scale the platform to 10 academic medical centers and develop a broader system called xLR, aiming to create a shared ecosystem for clinical knowledge and medical education.

Asset Subtitle

Snehal Bindra

Meta Tag

Author List Beau Hilton, Chase Jeffrey J. Webber DO FACP, Ivo Su, Joshua Onyango MD MEd, Matthew Sweeney, Michael J. Neuss, Snehal Bindra

Category Innovations

Concept VLRChat

Concept Retrieval-Augmented Generation

Concept Vanderbilt Internal Medicine Housestaff Handbook

Concept Hallucination

Concept Clinical Question Answering

Distinguished Finalist

Presenter Organization Vanderbilt University School of Medicine

Presenting Author Snehal Bindra

Track Technology in Hospital Medicine

Keywords

VLRChat

clinical decision support

retrieval-augmented generation

Vanderbilt Internal Medicine Handbook

medical education

evidence-based responses

hallucination reduction

hyperlocal guidance

clinical question answering

academic medical centers

VLRChat

Retrieval-Augmented Generation

Vanderbilt Internal Medicine Housestaff Handbook

Hallucination

Clinical Question Answering