Shortening the distance between thinking, building, testing, and learning.
Problem: Families facing a new Type 1 diagnosis juggle glucose readings, carb math, and emergencies across disconnected tools.
What I built: An AI glucose risk engine that turns CGM noise into plain-English caregiver alerts. 92% accuracy across 10K+ readings. Started after my twin sister's diagnosis.
Result: Pitched at Berkeley's BASICS accelerator. Featured in UC Berkeley L&S News.
Problem: 128M people watched Bad Bunny's Super Bowl halftime. Google Translate gave them "Uncle asked me" for "Titi Me Preguntó." The music hits. The meaning doesn't.
What I built: A Spotify-connected lyric sync engine with contextual translations, slang dictionaries, and cultural annotations. Poetry stays poetry. Slang stays slang.
Result: 50+ songs across 4 albums. 2K+ users. Sub-50ms lookup latency.
Problem: US colleges generate 12.1M tons of move-out waste every year. Students throw away furniture that incoming students buy brand new three months later.
What I built: A hyperlocal peer-to-peer marketplace. List it, match it, hand it off within the 72-hour move-out window. No app download, no shipping.
Result: UC Berkeley pilot: 127 items listed, 84 handed off, 8.4 min avg match time, 66% conversion.
Problem: Kids with motor and cognitive disabilities were locked out of learning games built for able-bodied play.
What I built: OpenCV/MediaPipe hand-gesture tracking for an adaptive learning game. Difficulty and input adapt to each child. Designed with clinicians in a Mumbai speech clinic.
Result: A playable clinical tool that met children where their ability actually was.
A diabetes-management app inspired by my twin sister's Type 1 diagnosis. Unifies glucose tracking, carb guidance, emergency alerts, and family monitoring. Pitched to investors at Berkeley's BASICS accelerator.
