I'm a product leader in San Francisco, technical across hardware, software and infrastructure at scale.

I've worked on motor control systems, LiDAR traffic intelligence for safety infrastructure, ML for battery analytics, mobile EV charging, grid orchestration for distributed energy, and space awareness for sovereign environments. Now it's AI systems for strategic work.

The throughline: make the technology disappear, and absorb the complexity so the person using it doesn't have to.

✍️ Atelier - [agentic systems, LLM evaluation, statistics, CLI] - turns examples of the work you want into an AI skill, with rules you approve once that no model update can move. Every output is checked against them and anything invented is cut, so you get work in your voice, run after run, without reading each draft.

🧾 receipts - [AI safety, fact verification, evals, Python] - is a guard for AI agents that research the web: every claim carries a quote from a page that code fetched, or it gets cut. You can repeat what an agent tells you without redoing the research, and every cut is listed with its reason.

🚐 depot-twin - [machine learning, fleet management, simulation, UI design] - is a digital twin of a robotaxi depot on a real East Oakland parcel that tells you how many vehicles one site's grid connection can carry, what breaks first as the fleet grows, and which charging rules get the most from its power.

⚡ MotionSim - [power electronics, motor drives, simulation, datasheet extraction] - takes any vendor's power-transistor datasheet, your motor and your load, and tells you what the power stage loses and how hot it gets. You get a part choice you can check, with the equations and the data behind every number.

🕯️ Epitaph - [generative art, edge AI, embedded systems, Raspberry Pi] - is an art installation where a language model runs on a small computer that I take away from it piece by piece while it's still thinking, until it dies and a new one is born. You get to watch how much of what a mind says comes from the mind, and how much from the machine shrinking around it.

🧰 skills - [AI skills, product strategy, go-to-market, Claude Code] - is a set of AI skills for Claude Code covering product marketing, strategy and go-to-market. You get an agent that does that work the way a senior operator would, from the first run.

🧭 StratOS - [AI agents, product strategy, go-to-market, SaaS] - helps early-stage teams decide who to sell to, what to say, what to charge and what to build next. Their agents read those decisions before building anything. Each decision spells out what would prove it wrong, and StratOS flags it when that happens.

Models are getting cheaper at everything objective. What stays scarce is the part that was always hardest to write down: what to emphasize, what to ignore, what feels wrong despite looking reasonable, which technically correct answer you'd never ship.

Agents now produce more than I can read. They can't own the outcome, and my attention doesn't scale with how many of them I run. My part is deciding what good looks like and checking that the work meets it.

Hardware taught me you can't hide behind iteration. You have to understand a thing all the way down to build it simply.