Case study
A decade of archives, five minutes from any answer.
Michael Thompson was sitting on ten years of his own best thinking. The problem was reaching it.
5 min
to surface what used to take 40
100+
transcripts indexed with attribution
Engagement: Custom AI workspace build
How do you train AI on your own expertise?
You don't fine-tune a model. You index your body of work with semantic embeddings, add a voice module trained on your writing patterns, and make every answer cite its source. The AI becomes an interface to your archive instead of a replacement for your thinking.
Who they were
Michael Thompson — entrepreneur and knowledge worker with a decade of accumulated material: 100+ transcripts, years of primary writing, and hundreds of books' worth of hand-curated quotes.
The problem
The archive existed. Finding anything in it didn't. A half-remembered idea meant hours of digging through folders and documents, and the search broke the creative flow it was supposed to feed. The work wasn't producing more material — it was reaching what already existed, ordering it, and opening it up to other people.
Scoping it honestly
The first conversation was about whether to build at all. For purely personal use, a standard AI subscription covers a lot — and I said so. Michael chose a custom build because he intends to productize the workspace. A builder who only ever says yes is a salesman.
What we built
- 01
Semantic index
Embeddings across the entire body of work, so search runs on meaning instead of keywords.
- 02
Voice module
Trained on Michael's writing patterns, so drafts and summaries come out sounding like him.
- 03
Reference library
Searchable and attributed — every answer points back to the transcript or text it came from.
- 04
Dual-panel workspace
Past conversations on one side, the assistant on the other. His history and his thinking partner in one view.
What changed
Research that took 40 minutes now takes five. Five distinct use cases emerged in the first weeks — summarization, idea testing, accessibility translation among them — and every one amplifies existing work rather than generating substitutes for it.
The archive stopped being storage and became an asset. Ten years of thinking, finally load-bearing.
Your body of work is the moat. The AI is the index and the mirror — never the author.
Your version of this
Every one of these started as a conversation.
Tell me what you're sitting on — expertise, archives, a content bottleneck — and I'll tell you what I'd build.
Not ready to talk yet? Start with the AI Readiness Diagnostic.