A useful clinical workflow
- The practice gives the patient the required information and records the consent decision.
- The consultation is captured on the clinic network and transcribed on the appliance.
- The Harness prepares the agreed note or letter template.
- The practitioner checks the recording context, transcript and draft, corrects errors and approves the final record.
- Only the approved output moves into the clinic’s system of record through the configured path.
Human review is part of the product boundary
Speech recognition can miss names, medicine terms, numbers, accents, speakers and negation. A fluent summary can still be wrong. The acceptance set should therefore include difficult audio, specialist vocabulary and examples where a small transcription error would change meaning.
Ahpra’s AI guidance says practitioners remain responsible, should apply human judgement, understand the tool, consider transparency and obtain informed consent where personal patient data is used. The OAIC’s guidance on commercially available AI explains that existing privacy duties continue to apply where personal information is involved.
Local is a narrower, testable claim
In the standard offline configuration, recording, transcript and model inference stay on the customer appliance. That can remove a hosted transcription provider from this path. It does not make the whole clinic offline: updates, web research, remote support and external practice-system connections are separate paths that must be recorded and assessed.
What the quote should name
- capture devices, rooms and approved users;
- consent prompt and how the response is recorded;
- templates, vocabulary and required review steps;
- retention and deletion for recordings and drafts;
- the system of record and any connected data path; and
- acceptance examples, error handling and support boundaries.
Read the companion guides on AI transcription and Australian privacy law and tracing where patient data goes.
Book a 20-minute walkthrough using non-identifying sample audio.
