Designing Intelligence.

I started as a musician. The discipline taught me that mastery is not an endpoint, it is the condition from which real expression begins. Twenty years of product design later, that idea still runs through everything I make.

I have designed for drivers at speed, for listeners in the dark, for engineers navigating machine intelligence. At Google, I built internal tooling for the first Gemini launch and earned a named contribution in the Gemini 1.0 technical paper. I now work on AI data infrastructure and tooling.

That proximity has sharpened one conviction above all others: the gap between what AI can do and what people will trust it to do is the most important design problem of our time. It will not be closed by better models alone. It will be closed by designers who understand both the technical and human sides, and possess the craft to build the bridge between them.

That is the work. That is what I do.

Four Principles for AI Product Design

Probabilistic UX

Certainty is a UX convention. AI operates differently. Most interfaces were designed for deterministic software where input goes in and predictable output comes out. AI breaks that contract. I design systems that communicate confidence levels honestly, handle ambiguity gracefully, and keep users oriented when the system isn't certain. The result is interfaces that don't erode trust every time the AI is less than perfect, which is often.

Impact: higher user retention in AI-powered features

Human-AI collaboration

The best AI products make humans more powerful, not more passive. Automation that removes human judgment doesn't scale. It creates liability and resistance. I design experiences where AI amplifies expertise: surfacing the right information, making recommendations legible, and keeping humans in the loop where they add the most value. Teams adopt it. Organizations trust it.

Impact: faster adoption across enterprise and consumer products

Invisible guardrails

Safety that interrupts is safety that gets bypassed. In generative AI products, constraints imposed as blockers breed workarounds. I build ethical guardrails into the experience architecture itself using interaction patterns that naturally guide users toward better outcomes without breaking flow. Safety becomes a design quality, not a compliance checkbox bolted on after launch.

Impact: reduced misuse without sacrificing engagement

Latency as interface

The moment between input and output is a design canvas. AI systems think before they respond. That latency is real and unavoidable, but how it's designed determines whether users experience it as broken or brilliant. I turn wait states into transparency moments: showing what the system is doing, why it's taking time, and what comes next. Done right, latency becomes part of the trust model, not a hole in it.

Impact: perceived performance gains without touching infrastructure