Chata-gata
Ask about my work, with sources.
The AI on this site. It answers from the portfolio itself and shows the pages each answer came from, so you can open one and check. It finds the right page first, without a model, so it still points you somewhere useful when the model can’t be reached.
Decisions and guardrails
Source before answer.
The page an answer came from shows first, so you can check it.
Retrieval first.
It finds the right page by searching the site itself, so it always points somewhere real.
Answers only from this site.
If something isn’t in the portfolio, it says so. Anything off topic gets a friendly no.
Built not to oversell me.
The biggest risk here is an AI talking up my work to a recruiter. So it keeps every caveat, and a concept stays a concept.
Limits outside the model.
It reads its own copy of this site, only this site can call it, and each visitor gets a few questions a minute. If the model fails, you still get my written answers.
Tested, misses included.
I asked it 50 tricky questions and read every answer. Early on it sounded stiff and named pages that don’t exist. It now passes every honesty check, but it can still slip.
Inside the app