YouTube Transcript & Summary

Purpose

Automatically extract transcripts from YouTube videos, summarize them using AI, and store both the transcript and summary along with metadata. Optimized for longer content.

Trigger

Manual form submission with YouTube video URL

Workflow

YouTube Video URL (Form Submission)
↓
Extract Transcript (Web Scraping)
↓
Normalize/Chunk Transcript (JS)
↓
AI Summarization (Gemini LLM)
↓
Store Original Transcript + Summary + Metadata (Airtable)
↓
Optional Notification
├─ Slack
└─ Email

Steps

  1. Trigger: User submits a YouTube video URL via a form
  2. Extract transcript using web-scraping
  3. Normalize and chunk transcript for AI processing
  4. Generate detailed summary using Gemini LLM
  5. Store original transcript, summary, and metadata in Airtable (e.g., video title, creator name, YouTube link)
  6. Optional: Send notification to Slack or Email

Notes

  • Currently manual URL input; future enhancement could include automatic detection of new videos from channels or playlists
  • AI summarization can be adjusted for level of detail
  • Works best on longer videos due to chunked summarization (transcripts are split for AI processing)
  • Respect LLM limits by keeping inputs within the model’s token capacity
  • Chunks can be combined if the LLM can handle larger inputs, allowing longer transcript sections to be summarized at once
  • Optional notifications can be added or customized
  • Only English transcripts are supported