Selected Work
Three players, one table, different tools. Dynamic tool registration gives each role different capabilities and data through the standard navigator.modelContext API.
W3C WebMCP standard · Chrome 146 native · Gemma 3 reasoning
View on GitHub →
Browser-native tool for exploring Waymo, nuScenes, and Argoverse 2. Streams LiDAR point clouds and camera views directly from local files or remote URLs, no Python or server needed.
3 major AV datasets · Remote URL streaming · Shareable scene links · Used by AV Triage (Encord)
View on GitHub →
Built a WebGL-based 360° image/video viewer from scratch with custom motion sensor fusion.
531★ GitHub · Amazon.com & Hyundai adopted · 1 patent granted
View case studies →
Systematic React profiling and optimization across 3 AWS console products.
97% render ↓ · 41% bundle ↓ · 89% API calls ↓
View case studies →
Redesigned model onboarding from 3-week custom code to config-driven system.
100+ models · Zero-regression re:Invent launch
View case studies →
Voice-driven programming game for kids and layered alt text for screen reader users. Two full-stack prototypes published at top venue.
2× ACM CHI publications
View case studies →Research & Patents
Co-authored study enhancing spatial understanding for screen readers with layered image descriptions.
Designed and built a full-stack prototype integrating voice UI and machine translation for programming education.
360° image object detection with radar-style bird's-eye minimap. Originated from NAVER Hackday 1st place win.
Server-side pipeline for auto-generating shareable content from uploaded photos and metadata.