Skip to content
All case studies

Healthcare · Enterprise AI · Multi-site clinic network

Ambient clinical notes that gave doctors their evenings back

An ambient scribe listens to the visit and drafts a structured note the clinician approves in one tap — written straight into the EHR.

ReactWebRTCWhisper (fine-tuned)Epic FHIR
Healthcare case study — Ambient clinical notes that gave doctors their evenings back

-40%

charting time

Timeline

5 months

Team

4 engineers + 2 ML engineers

The challenge

What was breaking

  • 01Clinicians spent two hours a night charting, driving burnout.
  • 02Hand-typed notes were inconsistent and sometimes incomplete.
  • 03Any automation had to be HIPAA-clean and clinician-controlled.

The approach

How we tackled it

  • 01Captured ambient audio with on-device consent and encryption.
  • 02Fine-tuned transcription and structured extraction on de-identified notes.
  • 03Kept a human in the loop: the model drafts, the clinician approves.

The solution

What we shipped

  • 01A real-time draft note that fills as the visit progresses.
  • 02One-click approval that writes structured data to the EHR.
  • 03Full audit trail and zero PHI leaving the secure boundary.

In pictures

A closer look

The ambient scribe in room
The ambient scribe in room
Extraction model
Extraction model
On-device PHI boundary
On-device PHI boundary

The results

-40%

charting time

+2.1

visits per day

96%

note accuracy

0

PHI incidents

"I look at my patients again instead of my screen. The note is waiting for me, not the other way around."
Attending Physician, Multi-site clinic network

Related work