Whisper and Pyannote via Replicate to spin up a container for transcription and diarization. It returns word-level timestamps and speaker turns.

The LLM pipeline is built into the Rust code and makes calls to the OpenAI API.