AI product designer and design engineer. I design AI that people can trust, and I build it myself.
I came to this from behavioral science. Four years in behavioral therapy, crisis counseling, and clinical support taught me something that turned out to be a design principle: people in distress can describe what is happening to them, but they cannot reliably pick the thing that would help. Most wellness software gets this backwards and hands you a menu at the exact moment choosing is hardest.
Now I build systems that do the selection, show their reasoning, and ask permission before acting. Currently finishing an M.A. in Human Interaction and Artificial Intelligence at Elisava Barcelona.
AuraPod AI · case study A multi-sensory regulation booth. A spoken or typed check-in is read zero-shot by Gemini onto the PAD affect model, then a deterministic weighted-Euclidean resolver maps those coordinates to one of ten regulatory states and runs five minutes of coordinated light, spatial sound, breathing, and scent.
Built solo end to end: research, affect model, prompt architecture, backend, frontend, hardware, design system, evaluation protocol. Self-taught Python and FastAPI over the course of it. 160 tests. Grounded in a systematic review of 104 peer-reviewed sources, and run live at the Elisava graduate exhibition with 40+ first-time visitors.
The design problem underneath it: a language model that misreads you is normally a bad autocomplete, but this one acts on your nervous system, so a wrong read is a breach. The estimate is probabilistic. What happens next is not.
crown-focus-logger Records focus, calm, and five-band EEG from a Neurosity Crown to CSV, with a mock mode so the pipeline is testable without hardware. The data-collection floor for ongoing work on whether consumer EEG can inform adaptive interfaces.
I build what I design. Prototypes that run on real hardware in front of real people, not mockups. I write tests because a system that acts on someone's nervous system should fail loudly and predictably, and I document what a project does not do as carefully as what it does.
I use AI tooling heavily, directed and verified rather than autonomous.
Python FastAPI WebSockets JavaScript Tailwind Gemini API pytest
Figma Framer Webflow
Zero-shot classification, system prompt design, human-in-the-loop UX, semi-structured interviews, qualitative coding, literature review, design systems.