100+ care sites
Sayvant documentation workflows across more than 1.1M charts.
Andrew Napier, MD
Clinical AI and airway tools, built by someone who still works in the ED.
I'm an emergency physician, former Army combat medic, and founder. At Sayvant: clinical AI documentation across 100+ care sites and more than 1.1M charts. At IntuBlade: a $95 single-use USB-C video laryngoscope used by EMS agencies across 41 states.
My lane starts when the demo ends: chart, airway, product call, 3 a.m.
1,089 QA iterations
Closed-loop production AI QA across 13 hospital sites and 42 physician complaints.
400+ EMS agencies
IntuBlade in 41 states, with 2,884 devices in field and $250K+ recorded revenue.
Army medic
Afghanistan service, 300+ documented casualties, Purple Heart, CMB, and CAB.
What I build
The work looks split on paper. It is one problem: get high-consequence tools to survive real humans, real constraints, and the parts nobody can hand-wave.
Clinical AI at Sayvant
I lead clinical AI product for emergency and hospital medicine documentation: physician-written rubrics, chart grounding, dangerous-fabrication review, PHI-aware routing, and post-launch QA.
Clinical AI portfolioAirway hardware at IntuBlade
I built a $95 single-use USB-C video laryngoscope because cost and logistics are clinical constraints. It's now used by EMS agencies across 41 states.
IntuBladeEvaluation and research
The public trail now includes a medRxiv QA preprint, ScribeBench, Sayvant SQS, Stanford AIMI chest pain work, and an AMIA 2026 accepted podium abstract.
Papers and proofEmergency medicine
I still practice in the East Bay. The bedside keeps the product work honest when a tool adds cleanup labor, false certainty, or one more box for a tired clinician to click.
BackgroundWhy anyone should care
A lot of health technology sounds fine in a slide deck. The hard part is what happens after it touches a shift, a chart, a budget, or an airway.
I still practice.
The ED is a useful filter. If a product makes the chart prettier but makes the shift worse, the clinician pays for it.
I have shipped.
Sayvant is in 100+ care sites. IntuBlade is in 400+ EMS agencies. Both products have customers, constraints, support tickets, and real consequences.
I measure failure.
The QA work tracked physician complaints through 1,089 iterations, reached zero regressions on resolved complaints, and eliminated 6/6 fabrication-class failures in ablation.
Start here
Pick the door that matches why you landed here. The formal record is available, but the clinical AI trail is the best entry point.
If you're here for clinical AI
Start with the clinical AI portfolio: production QA, model evaluation, ScribeBench, SQS, ED chest pain decision support, and implementation notes.
View clinical AI workIf you need the formal record
The CV has training, roles, publications, patents, awards, military service, and the ATS-safe resume artifacts.
View CVIf you want the project trail
Projects include clinical AI benchmarks, chest pain analytics, ICD-10 coding work, airway guidance, and public technical artifacts.
View projectsIf you want the short version
Read the writing. Most posts are notes from the collision between clinical reality, software, devices, and cost pressure.
Read writingRecent writing
All writingNotes on clinical AI, airway devices, documentation, and what changes when a product has to survive real care.
- - 5 min read
Preventing AI Fabrication in Clinical Documentation
One invented medication is enough. Clinical documentation AI has to be built around evidence.
- - 7 min read
Building a Chest Pain Risk Stratifier from Clinical Conversations
Chest pain exposes overconfident clinical AI. Useful support beats false diagnosis.
- - 5 min read
Changing Models in Clinical Documentation Without Losing Control
A new model is not a reason to touch the chart. Failure modes decide whether migration is worth it.
Papers and proof
All papersThe public research record is catching up to the operating work: production QA, chest pain decision support, airway hardware, and documentation fidelity.
Closed-Loop Quality Assurance for Production Clinical AI Documentation
medRxiv. 2026. DOI 10.64898/2026.05.27.26353977v1
Training Clinical Decision Support from Ambient Conversation
Stanford AIMI / Health AI Week and AMIA 2026. 2026. AIMI poster. AMIA podium abstract #15227 accepted
Decreased Time to Intubation with a Lens-Clearing Video Laryngoscope
American Journal of Emergency Medicine. 2021. PMID 33632548
Additional clinical publications available on request
Clinical publications and abstracts. StatPearls, Cureus, emergency ultrasound, and ACEP abstract work
Elsewhere at Sayvant
Content hubCritical Care Billing for EM Clinicians
Sayvant. January 27, 2025
Bridging the Gap Between Decisions and Documentation
Sayvant. December 2, 2024
Why "Defensive" Documentation Fails
Sayvant. October 27, 2025
Document trail
The CV has the formal record: training, roles, publications, patents, awards, military service, and the rest.