Current clinical AI work first. Older publications below that.
The filter is simple: did the work come from a real clinical problem, and can someone inspect the evidence trail?
These are the current outputs most relevant to clinical AI roles: production QA, decision support, documentation fidelity, and physician-grounded evaluation.
medRxiv · 2026
Preprint posted May 29, 2026. Andrew Napier, Justin Wiley, Mark Heslin. DOI: 10.64898/2026.05.27.26353977v1.
Artifact: first-author medRxiv preprint. Method: closed-loop QA using physician complaints against production AI documentation. Evidence: 13 hospital sites, 42 complaints, and 1,089 optimization iterations.
Stanford AIMI / AMIA 2026 / manuscript in preparation · 2026
AIMI poster, Stanford, CA, June 3, 2026. AMIA 2026 abstract #15227 accepted. Authors: Andrew Napier, Ashley Griffin, Mark Musen.
Method: paired ambient encounter signal with physician-authored decision traces, so the model is not trained only on downstream chart text. Status: manuscript and validation plan in preparation.
Public benchmark artifact · 2026
Clinical documentation fidelity benchmark for narrative quality, source fidelity, leak detection, and dangerous fabrication.
The benchmark makes the hard distinction visible: did the system preserve delivered care, or did it write something that sounds right and is not true?
These are older, practical clinical publications and book-length work.
American Journal of Emergency Medicine · 2021
Technical publication related to lens-clearing laryngoscopy. PMID: 33632548.
This came from a device problem I kept seeing: the camera view fails when blood, vomit, or secretions hit the blade. Theory is cleaner. The airway is not.
Cureus · 2021
Case report on transient reactive phlebitis involving vancomycin infusion therapy. PMID: 34754636. DOI: 10.7759/cureus.18471.
A weird ED reaction, written up cleanly. Not glamorous. Useful because unusual adverse events still have to be recognized on shift.
Independently published · 2020
Pocket reference guide for emergency medicine ultrasound.
I like compact tools that are actually useful on shift. This was built in that spirit: practical, quick, and meant to be used rather than admired.
Recent talks and founder presentations that connect the research to clinical operations and device work.
HITLAB Spring Summit · 2026
Sayvant session, May 5, 2026.
Implementation talk on what it takes to deploy AI in acute care delivery with health-system partners.
Stanford Emergence Annual Impact Summit · 2026
Stanford, CA, June 5, 2026.
Founder presentation for IntuBlade after the product reached commercial launch and expanded into procedural guidance research.
Eastern Kentucky University Emergency Medical Care Program · 2026
Guest lecture, Richmond, KY, April 2026.
Guest lecture tied to airway training, EMS education, and the IntuBlade device donation to EKU.
These are manuscript lanes still in motion. Some stay private until the manuscript, preprint, or benchmark path is cleaner.
Active preprint staging with Serge Klement and Ben Fedeles. The package uses connected video laryngoscopy, computer vision observations, event telemetry, generated reports, JSON exports, and trainer-confirmed safety fields.
Active manuscript and budget-impact appendix buildout for pediatric video laryngoscopy. The work ties device cost, first-pass success assumptions, downstream complications, and adoption economics together.
Active research/software work using California HCAI frequency data, synthetic EM/HM examples, frequency-aware retrieval, constrained LLM selection, and error analysis for a large clinical label space.
Framework and benchmark artifacts for structured quality scoring in clinical AI documentation. Current publication path is methods paper, appendix to the production QA work, or both.
Stanford bioethics and health policy paper with publication path pending after faculty review.