Patricio Valdez · Monterrey → San Francisco

Hi, I'm Patricio, friends call me Pato. I build production-grade AI agents, and I'm obsessed with human behavior.

I like building agents that are actually ready for production, not demos. Most of my work is the unglamorous part: evals, prompt work, and getting an agent from "works on my machine" to something a real customer depends on. I care about the math and the reasoning behind why something works, not just that it does.

We're early in the era of personal AI, and that's the part I can't stop thinking about. I'm experimenting at the edges of it: modeling how people decide, a memory layer for agents, agents that improve themselves, agents private enough and simple enough that my grandma could use one. Small bets in a direction I believe in.

Work

Seals AI→YC S24 · founding engineer

I build, evaluate, ship, and deploy the AI agents. Operational AI for LATAM distributors and wholesalers.

selfevals→open source

An evals framework that turns agents into self-improving ones, closing the loop so they get better on their own. It's how I cross the pilot-to-production chasm, at work and on side projects.

humanbehavior-predictor→open source

A model of how people make decisions and how those choices shape what comes next. One front of a bigger interest in human behavior. Writeup coming.

Non Negotiables→App Store

A habit tracker built on one idea: habits that log themselves. Plus group streaks and team activities. Hit #16 in Lifestyle on Mexico's App Store at launch.

brain_os→memory for agents

building A memory framework for AI agents: how an agent stores, recalls, and consolidates what it knows over time.

Guardian Angel→personal agent