- Status
- active
- Category
- technical
- Repository
- GitHub ↗
macOS / Apple Silicon pipeline that turns YouTube videos into structured Markdown summaries — from channel discovery through on-device transcription to optional Google Drive sync.
Problem
YouTube research produces long-form video; the useful signal is compressed, searchable notes. Manual transcription and summarization do not scale.
Core workflow
- Discover videos via channel crawl (YouTube Data API) or local input queue
- Download subtitles or audio (yt-dlp)
- Fall back to on-device Whisper MLX when subtitles are unavailable
- Preprocess transcripts with a lightweight LLM
- Summarize into mobile-friendly Markdown
- Optionally sync finalized output to Google Drive Desktop
What is distinctive
- Apple Silicon native — MLX Whisper for local STT without cloud upload
- Two-worker orchestration with single-writer shared state
- Transcript cache and per-video job workspace for resumable runs
- INPUT_PROMPT v3 structured summarization contract
Status
Active — GitHub.