The first version should accept uploaded or pasted transcripts, meeting notes, screenshots and recording links. A later connector can retrieve Teams transcripts and recordings through Microsoft Graph when the customer grants the required Microsoft permissions.
#Video and audio
For local processing, a customer-controlled worker can extract audio and run an engine such as faster-whisper. This avoids sending the recording to a hosted transcription provider, but the customer still needs adequate CPU/GPU, disk space and consent from meeting participants.
For cloud transcription, the customer chooses the provider and accepts its terms. AutoHub should record the provider, language, confidence metadata and retention decision.
#OCR and screens
Screenshots or selected video frames can be passed through OCR. OCR is evidence, not a selector. The Scribe should label observations as “seen”, “said”, “inferred” or “unresolved” so a reviewer can tell facts from model interpretation.
#Privacy rule
Do not upload a recording containing customer secrets, personal data or credentials unless the customer has approved that processing path. Prefer local processing for sensitive sessions.