Feedback on your consultations, without the cloud.
MedAdvisor records a patient consultation, transcribes it, and scores it against a medical educator's rubric — recording, transcription, redaction, and scoring all happen on the device. No audio and no transcript ever leave the phone.
In private pilot with a teaching program. Not yet publicly available.
How it works
1 · Record
Record a real or role-played consultation, with patient consent confirmed before capture begins.
2 · Transcribe
Apple's on-device speech engine produces a live transcript. Each utterance is then labelled clinician or patient.
3 · Score
A 7B language model applies the educator's rubric criterion by criterion, quoting the evidence for every judgement.
4 · Share, if you choose
Scores and redacted quotes can be shared with a supervising physician — only after you review the exact payload and confirm.
The privacy model
Clinical consultations are among the most sensitive recordings a person can make. The architecture starts from that rather than adding privacy afterwards.
- The AI never leaves the phone. Transcription and scoring run locally; the app works in airplane mode.
- Audio is deleted after analysis, and any recording orphaned by a crash is cleared on next launch.
- Transcripts are never uploaded. There is no field in the API that could carry one.
- Sharing is opt-in and reviewable. Only per-criterion scores and redacted evidence quotes can be shared, and only after the trainee reviews the exact content.
- You can take it back. A trainee can see everything their mentor holds and remove it at any time.
The rubric is the source of clinical truth — not the model. The educator's mark scheme defines what good looks like; the language model only applies it, and must quote transcript evidence for anything it credits.
Engineering notes
We write up the measurements behind the decisions — including the ones that went against the benchmark.