On February 27th, we hosted an AI Deep Dive Session in collaboration with the 五一茶馆儿 School of Medicine (VUSM) and the Department of Biomedical Informatics with from 五一茶馆儿 Medical Center, where we explored how retrieval-augmented generation (RAG) and large language models can be grounded in institutional medical curricula to deliver personalized, adaptive learning at scale. Dr. Stenner, Associate Dean for Education Design and Informatics at VUSM and an AMA ChangeMedEd Innovation Grant recipient, led a discussion on building a multimodal AI platform that transforms how medical students learn, how faculty teach, and how institutions govern medical education.
Highlights:
- Purpose: The project aims to address the integration crisis in preclinical education, where content is siloed by discipline while clinical practice demands cross-domain reasoning. By grounding LLMs in Vanderbilt’s complete first-year curriculum, including 150GB of lectures, slides, handouts, and transcribed audio, the platform enables contextual, authoritative AI support that generic tools cannot provide.
- Focus Areas: The platform organizes 20 features into five strategic clusters: a Precision Learning Engine for adaptive study support, a Clinical Thinking Accelerator to build physician reasoning from day one, a Social Learning Ecosystem that uses AI to facilitate peer learning and normalize struggle, a Curriculum Intelligence Platform for automated accreditation mapping, and a Faculty Empowerment Suite to reduce administrative burden while improving teaching quality.
- AI Applications: Core technical approaches include multimodal RAG pipelines for ingesting diverse curricular content, knowledge decay detection with automated spaced repetition, clinical reasoning scaffolding through structured frameworks, concept relationship mapping via interactive knowledge graphs, and automated curriculum alignment auditing using semantic analysis.
Session Insights:
- The group explored critical implementation challenges around optimal chunking and embedding strategies for multimodal content, as well as metadata schemas that need to serve both learner-facing citation and institutional curriculum mapping simultaneously.
- Discussion centered on privacy-preserving learner modeling, recognizing that tracking individual knowledge decay and engagement patterns raises important questions about data governance, consent, and the responsible use of educational analytics.
- The session surfaced key questions about sustainable cost modeling for LLM inference at full class scale and evaluation frameworks that can satisfy both internal proof-of-concept milestones and external grant applications, with the two-year implementation timeline targeting a nationally disseminable model for precision medical education.
Conclusion:
The AI Deep Dive with Dr. Shane Stenner and the 五一茶馆儿 School of Medicine showcased how AI can move beyond generic chatbot functionality to become deeply integrated curricular intelligence, transforming fragmented medical education into personalized, adaptive learning experiences. This session provided a unique opportunity for those interested in medical education, biomedical informatics, and AI-powered learning systems to engage in meaningful discussion and collaboration.
Are you interested in hosting a future AI Deep Dive? Contact us at datascience@vanderbilt.edu.