From a personal tool to editorial intake
The earlier TranscriptFlow experiment sent YouTube transcripts to Notion. The current implementation lives in the KSNK Media publishing workflow as youtube-postgres.
The current workflow
- I select a YouTube URL for processing.
- The service preserves the original audio and prepares it for speech recognition.
- Transcription jobs retain their responses and processing history, so a saved result does not need to be recognised again.
- Verified source material is registered in the Hub editorial store, with its files and hashes.
A transcript is source material. It does not automatically become an approved article, a translation or a publication.
What changed
PostgreSQL and source bundles replaced the earlier Notion-based route. The current design puts preservation, traceability and recovery before automatic rewriting. Ambiguous outcomes are held for review instead of being retried blindly.