YouTube batch transcription & processing pipeline

Eight nodes. One promise: any public YouTube video, in seconds, at any scale.

Pipeline nodes

1. 01
   ingest

   Accept URL, CSV, RSS feed, or channel handle. Deduplicate on ingest.

2. 02
   auth

   OAuth 2.0 into private/unlisted content where the caller has access.

3. 03
   resolve

   Detect available caption tracks. Prefer manual > auto-generated.

4. 04
   fetch

   Pull raw TimedText XML through pooled connections with retry + backoff.

5. 05
   parse

   Normalize to typed segments: text, start, duration, speaker.

6. 06
   enhance

   Optional: punctuation, translation, sponsor removal, LLM cleanup.

7. 07
   package

   Emit SRT / VTT / TXT / JSON, zipped per job.

8. 08
   deliver

   Webhook, S3, or download link. Signed URLs expire in 24h.