AI processing and practice data

Saras sends the minimum content needed for a requested feature to configured AI providers, which may include OpenAI or Anthropic. This can include written answers, OSCE transcript text, station context, and audio submitted for transcription or an enabled realtime interaction.

Saras does not intentionally persist raw recorded audio in its application database after transcription. Transcript events, candidate answers, marking results, and limited technical metadata may be retained so you can recover work, review feedback, and so Saras can operate and secure the service.

Production candidate data is not part of model evaluation fixtures unless separate approval has been obtained. Evaluation fixtures are synthetic and contain no candidate identity. Do not upload identifiable patient information or confidential clinical records.

Retention periods are configurable by the service operator. You can remove stored AI practice data from Settings. Account, billing, fraud-prevention, and legally required records are handled separately and may need to be retained.

To ask a privacy question, request help with deletion, or report a concern, email the support address shown on the site. We will verify the account request before changing or disclosing data.