Healthcare
Care note assistant
Turns a doctor–patient conversation into a structured visit summary for the doctor to review, removing names and other personal details first.
01 Problem
Clinicians spend a large share of each visit on documentation, and sending raw transcripts to an AI service risks exposing patient identity.
02 Approach
- Names, dates, phone numbers and other HIPAA Safe Harbor identifiers are replaced with placeholders before anything is stored or sent to the model.
- The model fills a strict SOAP note schema and lists what the visit did not cover instead of inventing it; a clinician reviews and locks every note.
- Finished notes export as FHIR DocumentReference resources.
03 Evidence
- Redaction covered by targeted unit tests
- Schema-validated model output
- End-to-end tests in CI
The repository README explains how to run each test suite, and its continuous integration history is public.