Ground the output
If a claim cannot be traced to source signal or the chart, it should not enter the record.
Andrew Napier, MD, FAAEM · East Bay, California · What I'm doing now
About
Before medical school I served as an Army combat medic and deployed to Afghanistan. That still shapes how I think about medicine and product design: systems have to keep working after the clean demo ends.
I now practice emergency medicine in the East Bay while building clinical AI and airway tools. Staying close to real patients, departments, and physician cleanup work is part of the product method.
The throughline
At Sayvant, I lead clinical AI for emergency and hospital medicine documentation. The work is less about making models sound smart and more about keeping them inside the record, preserving uncertainty, and making failures inspectable by physicians.
At IntuBlade, I built a single-use USB-C video laryngoscope around a different constraint: video airway management works, but cost, logistics, and device availability keep it out of too many ambulances.
Both efforts now have a public research trail—production documentation QA, ScribeBench, acute-care decision support, and video-linked airway telemetry—with deliberately bounded claims and reproducible evidence.
Jiu jitsu. White belt, training in the East Bay, and always up for a new partner. I built Jits Analytics, a training journal on web and iOS, because I wanted to see my own progress between sessions.
Veterans. Army combat medic, Tillman Scholar, and glad to talk with any medic or corpsman thinking about medicine or building something.
Building for fun. 3D printing, CAD, and small software tools, usually built with AI agents, usually late at night.
Field record
Led battalion-level medical operations and treated more than 300 documented casualties during a deployment to Afghanistan.
Trained in emergency medicine, then led quality, throughput, documentation, and sepsis improvement work in a 70,000-visit department.
Turned an airway problem into patented, FDA-regulated Class I hardware used by 400+ EMS agencies across 41 states.
Leads clinical AI product and evaluation work for documentation systems used across 100+ care sites and more than 1.1M charts.
Completed the MCiM program with work in physician-grounded evaluation, acute-care decision support, and production AI quality.
Clinical AI
If a claim cannot be traced to source signal or the chart, it should not enter the record.
Clinicians need ambiguity represented honestly, not rounded into false confidence.
Evaluation, escalation, and post-launch monitoring are product surfaces—not cleanup tasks.
Contact