Built between roles. Same approach as everything else here: find the real problem, model the system, ship something that works. Just compressed into days instead of quarters.
These are that record. Each one starts with a real product problem, runs through structured object modeling and design system work, and ends with something you can actually click into and use, not a static mockup. Along the way I kept a running log of every place the AI got it wrong, because that’s the part that actually proves the process holds up.
They’re listed most recent first, and not by accident: each one pushed further than the last, from a weekend app to a full agentic product with a tested working implementation. If you only have time for one, start at the top.
Pick a project below to see the build, the mistakes, and the fixes.
Designing an AI Agent
What does an AI agent feel like to use, when there’s no pattern library to borrow from?
CampOps Consumer is a self-initiated POC exploring agentic product design: an AI travel agent that recommends, adapts, and asks permission before doing anything irreversible. Thirty-three screens, a working Next.js build, and eight tested scenarios later, it’s a real answer to what “legible delegation” looks like in an interface.
Figma Still Wins
The handoff gap developers feared was smaller than they thought.
An engineer once questioned whether design handoff still meant redoing a designer’s work by hand, since prototypes don’t translate into anything a dev team can use. This POC tests that fear directly. I built a full design system and five screens in Figma, then used AI tooling to carry them into production-shaped code matching a specific target stack. A side-by-side review found exactly one discrepancy across five screens. Figma wasn’t the bottleneck. Getting there also meant catching about 35 places where AI output drifted from intent, logged live as it happened, proof that the gap closes with direction, not by itself.
OOUX Design + Vibe Coding
From brief to deployed app in two days.
Giftify is a full gift-tracking app with real auth, a real database, and a working AI suggestion engine, built solo over a weekend by following a structured OOUX-to-AI-build pipeline start to finish. It’s the answer to “can you actually build things,” and the answer is live.
