n8n Transcript Course Analyzer

1. Start with Manual Trigger or Webhook input (video_url, video_id, analysis_goal).
2. Use HTTP Request to call transcript source.
3. Normalize rows into [start, duration, text] records.
4. Use Code node to join into full transcript and estimate token count.
5. Use Text Splitter / chunking logic at semantic boundaries (headings, timestamp breaks, sentence endings).
6. Send chunks to LLM node with strict JSON schema output:
   • module
   • concepts
   • n8n_nodes_mentioned
   • business_use_case
   • implementation_notes
7. Aggregate chunk summaries into final synthesis:
   • executive summary
   • key workflows taught
   • monetizable services
   • implementation checklist

Path B — Fallback Provider + HTML Extraction

1. Resolve video_id from URL in a Code node.
2. Request fallback transcript page (example pattern in references file).
3. Extract transcript blocks with HTML Extract or regex in Code.
4. Clean artifacts (timestamps, duplicate whitespace, UI text).
5. Continue with chunking and synthesis from Path A.

Path C — ASR Fallback

1. Download audio (or use existing audio URL).
2. Transcribe with speech-to-text provider.
3. Add confidence metadata and clear disclaimer (source=asr).
4. Continue through same chunk/analyze/synthesize pipeline.

Output Contract

Return a JSON object with:

• video_title
• summary
• key_points (array)
• workflow_patterns (array of {name, n8n_nodes, purpose, monetization_angle})
• action_plan_7_days
• risks_and_gaps
• transcript_source

Quality Rules

• Preserve original meaning; do not invent claims not grounded in transcript.
• Use timestamps when available for each major takeaway.
• Separate beginner setup steps from advanced agent orchestration steps.
• Flag all advice that depends on external credentials, pricing, or API limits.

Troubleshooting

• Empty transcript response: rotate provider or switch to ASR fallback.
• Over-token transcript: summarize per chunk, then recursively summarize summaries.
• Poor extraction quality: remove boilerplate lines before chunking.
• Inconsistent outputs: enforce JSON schema and retry with deterministic settings.