Prototype for ROSC · synthetic data
Model:
For ROSC · built by Truong Pham

Edit notes by voice. Never lose a clinical fact.

An AI assistant for the note workflow: providers shape templates by asking, edit notes by typing or speaking, and insert their own wording with a spoken shortcut. Every edit is checked against the clinical facts before it can be applied.

Try the safety checkHow it works

Templates by asking

“Add a patient education section after the plan.” The assistant proposes a structured change, previews it on a real transcript, and saves it as a new version you can roll back.

Edits that can't change the facts

Doses, vitals, allergies, negatives and follow-ups are extracted by code before and after every AI edit. Anything changed without being asked is blocked.

Spoken macros

Say “insert normal cardiac exam” while dictating and your saved wording appears, even when speech recognition gets a word slightly wrong.

How the safety check works

The AI drafts. Plain code decides whether the draft is safe to apply.

1 · ExtractClinical facts are pulled from the current note: medications with dose and frequency, vitals, results, allergies, negatives, follow-up.
2 · DraftThe AI edits the note from the provider's typed or spoken instruction.
3 · CompareFacts are extracted again and compared. A change counts as intended only if the instruction itself asked for it.
4 · DecideSafe edits show as a clear diff to apply. Unrequested changes are listed and Apply is locked.

Live demo

Synthetic encounters only. Start with “Generate note”, then “Simulate a faulty AI edit”.

Template

Versions

    Tell the assistant what to change