Safety principles

Clinical review: AI-generated documentation is reviewed by the clinician before finalisation.
Traceability: timestamps, workflow events and audit information are maintained across every session.
Data protection: sensitive health information is protected through access control, encryption, tenant isolation and secure operations.
Data minimisation: only information required for the intended workflow is processed and retained.
Model governance: AI models and supporting components are versioned, evaluated and monitored.

What we evaluate

Transcription accuracy
Speaker attribution
Clinical entity recognition
Omission and hallucination rates
Medication and dosage preservation
Negation and temporal handling
Note completeness
Specialty-specific performance
Multilingual performance where offered
Template adherence
Regression after model changes

How we publish performance claims

Every performance metric MedTalk publishes states its methodology, the population it was measured against, sample size, date and known limitations. See the Research & Validation page for published evidence.