I pointed an AI pipeline at 200 golf instruction videos and turned them into a single queryable expert. Here's how.
A Python script scraped the full channel catalog — every video ID, title, date, and duration — and wrote it to a structured manifest. No API key needed.
Fetched English transcripts for the 200 most recent videos. YouTube blocked our IP after ~23 requests. Fix: VPN + randomized 5–30 second delays.
Claude read every transcript and wrote a structured summary: topics, key takeaways, drills, contrarian positions, and Shawn's signature analogies. Batched in groups of ~20.
Pass 1: extracted 210 unique tags and grouped them into 19 teaching buckets. Pass 2: deep-dived each bucket by re-reading every relevant summary. Pass 3: assembled the final 95KB knowledge base.
Loaded the knowledge base into a Claude Project with a custom system prompt. You can ask it anything about Shawn Clement's teaching and it cites specific videos.
A Claude Project loaded with 95KB of synthesized golf knowledge. Ask it what Shawn thinks about weight transfer, which drills fix a slice, or what videos to watch for putting — it cites the source.