I made my own personal AI business coach

I built a chatbot trained on 1,000+ videos from any YouTube creator.

Here's the workflow.

Claude just released "co-work", computer use for local files. But I've been doing this for months.

its just one of the processes:

Download subtitles from entire YouTube channel (not videos, just text)
Generate embeddings for all transcripts
RAG over them with local LLM

Result: Chatbot that answers like that creator

Example:

Want Alex Hormozi's business advice?

Download his 1,000 video subtitles.
Train chatbot. Ask business questions.

Want Dan Koe's content strategy? Same process.

I don't use Claude to process the data (saves tokens).
I use Claude to write the scripts that do the processing locally.

Claude = Blueprint generator.
My machine = Execution engine

The real challenge I'm exploring is this:
What happens when you combine MULTIPLE creators into one chatbot?

Single creator RAG is Simple 😎
Multi-creator synthesis is Hard 😅

Do you:

Fine-tune separate models and combine them (mixture of experts)?
RAG over everything and let LLM synthesize?
Fine-tune one model on all creators?

I'm testing RAG first. Cheapest to iterate.

While Most people use AI to consume content faster, I'm using AI to extract knowledge patterns from hundreds of hours of content I'd never watch.

That's the difference between tool user and tool builder.