Beak

Guide

How a real-time interview copilot turns speech into a useful prompt.

A live copilot is a pipeline, not a single model. Audio must be captured, transcribed, grouped into a question, combined with approved context, and sent to a language model before a suggestion can appear.

1. Audio becomes text

The desktop app opens separate microphone and computer-audio streams. A speech-to-text provider returns partial and final transcript segments.

  • Audio permission and capture
  • Real-time transcription
  • Speaker/source labeling

2. Recent context becomes a prompt

The service keeps a bounded recent transcript, the user’s direct request, and selected résumé or role material. Older or irrelevant text is excluded to control latency and cost.

3. The model drafts; the person decides

A capped model response is streamed to the overlay. The user checks it, adapts it, or ignores it. Transcription errors and incomplete source material can still produce a wrong suggestion.

Sources

Factual and competitor-plan claims use the sources below. Date checked for comparison sources: 30 July 2026.