While designing a feature that turns uploaded meeting recordings into speaker-labeled text, our decisions on scope, diarization model placement, upload method, and call order changed one after another. This post traces how.
Posts in this category are on the way
