        
        Clinical AI · Built & Examined
      

The clinical AI company that shows its work.

Celeritas builds AI for the places medicine is hardest to reach — and studies how anyone should test whether clinical AI is actually right. Two threads, one discipline: assume it fails, find where, and make the dangerous failure impossible or visible — starting with an AI that never states a drug dose.

Explore the work Read the research

Why we exist

You cannot credibly judge clinical AI unless you have built one and lived with its failure modes. So we do both — we build it to a standard, and we research how everyone else's should be tested.

What we believe

One engineering doctrine runs through everything Celeritas touches — the build and the research alike.

→ ASSUME FAILURE

Design for the failure mode first. The dangerous failure is a confident wrong answer.

→ MEASURE THE DANGER

Not "how often right" but "how often confidently wrong" — on tests a system can genuinely fail.

→ ENFORCE IN CODE

Safety-critical behavior is a deterministic code path, not a polite request to a model.

→ NO SELF-GRADING

Let no model grade its own work. Independent review, or it doesn't ship.

→ DOCUMENT LIMITS

Known blind spots are written down and tested as known failures — not discovered by the user.

Two lines of work

▸ BUILD · IN PRODUCTION

CodeBlue Peds

Pediatric protocol guidance over plain SMS — grounded in an institution's own validated protocols, PHI-stripped in code, independently reviewed, and engineered to fail safe. No app, no login. The interface is a phone number.

See how it works →

▸ RESEARCH · IN DEVELOPMENT

Testing clinical AI

An open exploration of a hard, unanswered question: how should anyone test whether a healthcare-AI system's clinical outputs are actually correct? Methods and preprints, developed openly — pediatrics first.

Read the research →

▸ NEXT · TBD

More in development

Celeritas is an umbrella for applied clinical-AI work. Additional threads will surface here as they mature.

Built by the specialist on the other end of the phone

Benson Hsu, MD, MBA, FAAP, FCCM

Founder, Celeritas · pediatric intensivist & professor

Board-certified in pediatrics and pediatric critical care, with 75+ publications and contributions to 10+ national practice guidelines — including the Surviving Sepsis Campaign children's guidelines. Former health-system VP of enterprise data & analytics leading a 70-member team across a $6B+ system. Princeton AB; Duke MBA; Harvard Kennedy School; former McKinsey. Bush Fellow, Aspen Health Innovators Fellow, Presidential Leadership Scholar.

"As a PICU attending in rural America, I've been the specialist physicians call at 2 AM when a child is deteriorating. I've taken thousands of those calls. This is what I wish every one of those physicians had in their pocket."

Let's talk

Building clinical AI, buying it, or want to help test it?

We're speaking with health systems, clinical-AI builders, and pediatric clinicians interested in the working group. No demos, no waitlist — just a conversation.

benson.hsu@celeritashealth.com Join the clinical working group