The hardest part of building an AI agent trained on my case studies wasn't really the wiring. It was deciding what it should refuse to say.

Ask it where I live and it tells you to ask me.

Ask for a metric it doesn't have and it says so, instead of inventing a number that sounds about right. That one took the longest. A model will happily produce "40% lift in task completion" because that is the shape of a sentence that belongs in a case study. Getting it to say "I don't have that" meant deciding, claim by claim, what it was allowed to assert.

Every AI feature has a version of this problem. The model is rarely the hard part. The hard part is the boundary: what it does when it doesn't know, and whether the person on the other side can tell the difference.

That's a design problem, not an engineering one.

Live on my portfolio if you want to try it: chrisjmartini.com/ask-ai-about-my-work