The KEV.AI Lab works across two interlocking research programs, united by a single commitment: making medical expertise computable and trustworthy.

Research Themes

1. AI for Medical Education

Engineering AI tutors, simulators, and competency-based assessment systems that transmit clinical expertise at scale — and validating that what learners acquire actually matches what experts know.

Core questions:

  • How can large language models be engineered into reliable, safe, pedagogically sound clinical tutors?
  • What evaluation frameworks move beyond “does it sound right” to “does it teach correctly”?
  • How do adaptive learning systems close competency gaps across the 17-million-strong healthcare workforce — not just physicians?

2. Clinical Knowledge Extraction & Validation

Building and validating AI pipelines that transform unstructured electronic health record data into research-grade evidence for registries, quality programs, and clinical trials.

Core questions:

  • How do we validate AI chart abstraction against expert human review at scale?
  • What ontologies and representational frameworks are needed for semantic interoperability in surgical and acute care?
  • How do we build auditable AI memory infrastructure for clinical trials and regulatory-grade research?

3. Responsible Deployment & Governance

Crosscutting both themes: the organizational, regulatory, and sociotechnical infrastructure required to deploy medical AI responsibly.

Active Projects

SchoolMe

An AI-powered adaptive learning platform for medical education, built on foundation models and licensed from VUMC. SchoolMe operationalizes the lab’s AI-in-education research as a commercial platform serving medical societies, health systems, and learners directly.

  • Current partnerships: American College of Surgeons educational program (in development); ChenMed PDV coaching simulator (active)
  • Platform: schoolme.me
  • Lab role: Origin, research partner, and primary translational outlet for AI-in-education work.

BRIM Analytics

AI-guided chart abstraction for clinical registries, funded by ARPA-H (Democratized, AI-Guided Chart Abstraction Platform, RSO-ISO-5011-P; $1.98M; PI: Daniel Fabbri). The BRIM platform is active across multiple clinical domains:

  • Urology — VUMC Combined Prostate Biopsy Database
  • Hernia surgery — ACHQC boolean variable validation
  • Biliary surgery — Wake Forest biliary registry collaboration
  • Organ procurement — Tennessee Donor Services abstraction
  • VUMC revenue cycle CDI pilot (malnutrition, sepsis, AKI)
  • ARPA-H workstreams on OKRA and DAGCAP

Lab role: Co-Investigator leadership on AI-guided chart abstraction for clinical registries.

TIPTOE — Trauma Institutional Priorities and Teams for Outcome Efficacy

NIH/NIGMS R01 GM111324 ($1.88M; PIs: Sexton, Brochhausen). Studies organizational and team-structure features that drive outcomes at Level I trauma centers. Embeds ontology development (Early Warning System Scores Ontology, CAFE trauma center data service) as core infrastructure.

Sociotechnical Framework for Risk Stratification in CHF

NIH NINR R21NR021063 ($407K; PI: Williams). Developing a sociotechnical framework for equitable care delivery in African Americans with congestive heart failure.

Health Sciences Innovation & Entrepreneurship (HSIE) Training Program

NIH/NCATS T32 TR004918 ($948K; PI: Sexton). A translational training program housed at UAMS, handed off to the next PI team as the lab’s center of gravity shifts to Vanderbilt.

Exploratory & Emerging Work

  • Rockefeller Bellagio Center EOI — Multimodal AI and global healthcare workforce training (submitted with Kyla Terhune, SVP Education, American College of Surgeons).
  • Auditable AI memory infrastructure — Joint exploration with amotivv on Memory Pod Fabric (MPF) for cryptographically auditable AI-assisted clinical trial enrollment, targeting the LITES Network.
  • Venous waveform monitoring — Continuing translational program on peripheral venous pressure for hemorrhage and dehydration detection; technology licensed to Baxter International.

Infrastructure & Tooling

We build and maintain a set of internal tools used across lab projects:

  • AI Literacy Scanning Skills — Automated pipelines (ai-edu-scan, healthcare-ai-scan) that surface weekly peer-reviewed and preprint literature across AI-in-education and clinical AI.
  • Granola-to-Obsidian Workflow — A PKM skill connecting meeting intelligence to longitudinal research notes.
  • ABLS AI Tutor Prompt Architecture — A structured, versioned approach to clinical tutor design (currently at v8).
  • Ontology assets — Including the Early Warning System Scores Ontology and the CAFE trauma center structural ontology.

Funding Partners

See the full list on the Publications and Partners pages.