Electron desktop app ships as one code-base for macOS, Windows and Linux. A tiny Swift CLI grabs lossless system audio + screenshots on macOS; Windows uses Chromium loopback.
Mic and speaker are captured as separate streams, so Google Speech-to-Text (Deepgram later) and GPT-4o can label the interviewer vs. the user in real-time.
UI suggestions arrive in < 500 ms.
Landing page lives on Bolt.new; demo videos were generated automatically with Claude Code.
Full build notes, diagrams and open-source snippets in the post.