Saco la IA del laboratorio y la llevo al trabajo de las personas.
I lead product strategy and operations at Google, taking AI features from their first internal build to general availability. Here is how I got here, plotted.
Now, Oct 2026: running alphas for agentic features, moving my tee-time watcher onto AWS, and modeling wind turbine failures from SCADA sensor data.
Seven chapters, one direction. Time runs left to right. Tap a stop to see each chapter.Time runs left to right. The vertical axis shows how much I owned in each chapter, or switch to AI depth to see how close the work moved to the models. Each cloud of points is the work in that chapter, and it grows as the work did. Pick a stop and its chapter appears just above.
Most recent first.
I take AI features from first internal use to general availability, side by side with early customers and the teams building for them. I write evals, prototype agents, and make sure every tester hears what happened to their feedback.
I earned a master's in analytics at Georgia Tech while working full time at Google. I wanted the math under the systems I was shipping: statistics, optimization, simulation and machine learning.
I ran programs for ML data quality across a global team. A model reflects the people and processes behind its data, so that is where I put my attention.
I joined Google working on what machine learning is made of: how data gets collected, labeled and checked before a model ever sees it.
I sat between complex software and the businesses running on it. Most problems turned out to be translation problems, and I got good at translating.
Where I learned that operations is a craft. A growing, multi-location business only works when a hundred small things happen on time, every day.
Where it started. Along the way I added a product management certificate from Stanford and a Lean Six Sigma Green Belt from Purdue, because I like knowing why a process works before I change it.
A few things I keep coming back to. HoverTap one to watch it come out of the noise.
Close the loop.
If you asked someone for feedback, tell them what happened to it.
Ship when the people using it say it's ready.
The calendar is a guess. Usage is evidence.
Automate the busywork. Keep the judgment.
The best AI tools give people back time to think.
Explain it in two sentences.
If I can't, the decision isn't done yet.
Short notes from taking AI features out of the lab and into real teams.
Early access programs usually fail quietly. Testers who never hear back stop testing honestly, and the next cohort is harder to fill.
So before anything I run reaches general availability, every tester gets a note: what we changed because of them, what we didn't, and why. It takes an afternoon. It turns a pilot into a relationship.
Agents fail differently than chat. The final answer can look right while the steps behind it are wrong.
When I run an alpha for an agentic feature, I write the eval rubric with the customer's toughest reviewer in mind. Did it take the action they would have taken? Would they let it run unsupervised? What would they need to see to say yes? A rubric built that way can turn a nervous customer into a committed one.
Most early access programs start as a spreadsheet and a favor. When I built one from scratch, I started with intake: one front door tied to the product roadmap, so every request maps to something a team is actually shipping.
Then I let AI draft the outreach to product managers. People spend their time deciding which customers fit. The model writes the emails.
I build small tools for myself with Claude Code, Codex and whatever else fits the job. A few I use all the time.
Good public tee times vanish in seconds, so latency matters. It deploys to an AWS instance in the same region as the booking system, or falls back to a Mac at home, with both nodes on a private Tailscale network. The moment a time opens, it moves.
A price-comparison project that tracks watch listings in Japan and the US side by side. Built with a friend who loves watches as much as I do.
An interactive 3D model I use to plan projects room by room. A friend saw it and now wants to build his own.
A small home lab that keeps my own tools running and connected, with ads blocked across every device.
Watches are the other obsession: Japanese and Swiss, mechanical and quartz, the history and the engineering. I think in English and Spanish, often in the same sentence, and play golf whenever the tee-time watcher comes through.
This one keeps New York time. Pick a movement and watch the seconds hand change how it moves.
Building something people need to trust? I'd like to hear about it.