“AI scribe” can sound like a black box. It shouldn’t be one. Here’s the actual sequence, no marketing language.

  1. Live audio streams straight to Amazon Transcribe Medical over an encrypted connection—it never touches Tasrif’s own servers. If you upload a recording instead of dictating live, that audio is held only long enough to run the transcription job, encrypted the whole time, then deleted the moment the job finishes or fails.
  2. A transcript is produced, then handed to Amazon Bedrock to draft a structured note in your clinic’s template. Bedrock doesn’t train its models on what you send it and doesn’t share it with other model providers.
  3. You review it. Nothing is filed until you’ve read it, corrected anything wrong, and confirmed it.
  4. What’s kept is encrypted individually—not just “the database is encrypted” as a blanket claim, but each patient’s name, each transcript, and each note, sealed with its own key managed through AWS’s key management service.
  5. It stays until your clinic decides otherwise. By default nothing is auto-deleted—clinical records often need to be kept for a minimum period anyway. If your clinic sets a retention period, or a patient asks for their record to be erased, we anonymize it rather than deleting it outright, so the rest of your records—and the audit trail of what happened—stay intact.

Every view, edit, and export along the way is logged—who, what, and when—and available for your own review. That’s the whole path. No step we didn’t want to describe.