Can a prompt replace MeetGeek?

Build a meeting-highlight extractor on top of local transcription, extending this catalogue's Avoma entry's approach with automatic clip generation. Use Python with FastAPI, whisper.cpp for local transcription, and ffmpeg for video processing. After transcribing an imported recording, explicit user import, not automatic bot-joining, same principle as the Avoma entry, scan the transcript for candidate highlight moments using simple, explainable heuristics: explicit flag phrases the user can configure, such as 'this is important' or 'action item', or a basic speaking-pace-change detector as a secondary signal. For each flagged moment, cut a short video clip, a configurable window, such as 20 seconds before to 10 seconds after, from the original recording using ffmpeg, with the matched phrase as the clip's caption. Present flagged clips in a review list where the user confirms or discards each before it's kept, since heuristic detection will produce false positives. Export confirmed clips as individual MP4 files. Do not build a more sophisticated ML-based highlight-detection model, calendar-triggered automatic recording, or team-shared libraries — those are out of scope; this uses simple, explainable heuristics rather than a trained model, and says so. Needs hosting to run continuously, or can run on a personal machine for occasional use; no external API key required.