Fair warning, these case studies run long. I document the research, the trade-offs, and the decisions I got wrong before I got them right, because that reasoning is the actual work. If you're short on time, the headline and metrics below each project will tell you most of the story.
My first product role wasn't an AI role. It was two and a half years turning a live, in-person property-investing masterclass into a paid online platform for adults with money on the line, running research in four countries, designing the interface, directing the video, and handing off to engineers. It worked, and it taught me what I was missing.
A live masterclass couldn't scale past the room it was taught in. I turned it into a self-paced platform for working adults learning to invest in property, 50,000+ paying learners in year one, and it still runs today.
I'd been running research on instinct. So I went to Iowa State for an MS in Human-Computer Interaction and chose my focus deliberately: safety-critical AI, human-AI trust, and multimodal interaction. Since then I've stopped drawing the line between designing a system and building it, every project below, I researched, designed, and shipped code for.
Live prototype, recorded on a Pixel 7 during testing. Flow: hazard detected → voice cue → haptic confirmation.
Electric vehicles are quiet enough to erase the cue blind pedestrians rely on to cross safely. I designed and built a voice-first system that sees the hazard and says so, on-device, in under half a second. Tested with 8 blind participants through the Iowa Department for the Blind.
Journalists and researchers who blur faces can't risk one miss. I designed a fail-safe architecture that hides every face by default and only reveals what's explicitly whitelisted, running entirely offline, on CPU, with nothing leaving the machine.
Managers, doctors and HR staff get one shot at hard conversations and no safe place to rehearse. Pathic puts you on the receiving end of your own words in mixed reality, with an AI counterpart that stays psychologically consistent rather than merely factual.
People were arriving two hours early just to secure a space. I designed a two-sided marketplace around predictability instead of search, then proved it, with a comparative study showing a statistically significant improvement over the baseline.
People care about their footprint but can't feel it, the consequence is decades away and the number is abstract. I designed for temporal proximity, turning distant impact into a choice you can see at the moment you make it.
Timing, feedback, and what a system does while you wait are decisions, and they only show up in motion. Two walkthroughs of prototypes I designed and built: one on desktop, one in the hand.
A full pass through the flow, states, transitions, and system feedback in real time.
Touch, motion, and feedback as they land in the hand.