Should you build your own Otter.ai?

Build me a personal meeting transcription and search tool to replace Otter.ai.
Requirements:

- Python stack: whisperX (faster-whisper backend) for transcription, Flask for the UI,
  stdlib sqlite3 for storage.
- A CLI: `otter record` captures the mic to ~/Meetings/YYYY-MM-DD-HHMM/audio.wav; `otter
  import file.m4a` handles recordings made elsewhere.
- Transcribe locally with whisperX, word timestamps plus speaker diarization; label
  speakers SPEAKER_1/2 and let me rename them once per meeting.
- Send the transcript to an LLM (key in .env) for a summary: 5 bullets, decisions made,
  action items with owners. Save transcript.md and summary.md next to the audio.
- Index transcripts into SQLite FTS5; `otter search "budget"` returns matching lines
  with meeting date and timestamp.
- A minimal page on localhost:8787: meeting list, one search box, and an ask box that
  answers questions over a chosen transcript via the LLM.
- Everything stays on my machine except the LLM calls; no accounts, no telemetry.
- Out of scope: a bot that joins Zoom/Meet calls, mobile apps, and team sharing.
  Diarization will be rough on crosstalk, accept it.
- README: Python and ffmpeg install, model download size, and the macOS mic permission.