01
EXPERIMENT
One SQL tool vs. one tool per system
An interactive demo where you replay every recorded run and compare three ways an AI agent can answer the same question.
View projectOpen to new engagementsoi! I’m Stefano.
judges rely on this. it has to be right.h 272I make AI
I make AI
behave
where
mistakes
are expensive.
Senior engineer and former head of product. I take AI from idea to production for courts, government and regulated enterprise.
the receipts
Three numbers I can back up.
COST
~10x
lower LLM cost per run
ZERO TO ONE
#1
employee at Vivacity, team grown to 10+
CRAFT
13
age I started writing code
Thoughts
Writing on this site, on LinkedIn and wherever else it lands.
short ones, mostly about evals.Projects
Things I built or led from zero.
the fun partWork with me
Consulting, invoiced through my own company.
pick one. or don’t, and just say hi.Prototype to production
It demos well but nobody trusts it yet. I build the evals and source checks, then cut the cost per run.
Fractional product and AI lead
For early teams that need one person on customer calls, the roadmap and the code. I did that for four years as employee #1.
Is this AI feature working?
A week or two reading your pipeline and your LLM bill. You get a written list of what to fix first.