50
Creators studied
6,479
Transcripts ingested
16
Parallel scrapers
Overview
Guy Guru is a corpus and agent platform I built from the ground up. It ingests thousands of hours of long-form creators, analyzes exactly how each one talks, and lets an AI write, advise, or roleplay in their voice.
The difference is grounding. Every output traces back to real transcripts with line-level citations, not vague imitation. It spans 50 creators and 6,479 transcripts, each backed by a detailed voice analysis of vocabulary, cadence, and signature frames.
What it does
- ▹50 creators studied, 6,479 transcripts ingested and counting.
- ▹A custom scraping pipeline (yt-dlp, 16 parallel workers) that gets past YouTube's caption gate.
- ▹Per-creator voice cards: vocabulary, cadence, signature frames, and anchor quotes with citations.
- ▹A loadable superagent that writes, advises, or runs cross-creator analysis on demand.
- ▹Every claim cited back to a real transcript line. Transcripts are the source of truth.
What I learned
“Retrieval grounding plus citation is what makes an LLM trustworthy enough for real work, not a bigger model. If the output cites its source, you can verify it instead of hoping.”