About
I’m an EECS student at UC Berkeley. Most of what I build starts as a question I want answered: what a defense forum has said about a jet over the years, how China’s industrial output compares with everyone else’s, how Shenzhen’s skyline compares to other cities.
The common thread is keeping the source attached. If a tool gives me an answer, I want to be able to click through to the post, report, or record it came from.
I’ve also done computer vision research at Boston University and UC Irvine.
Projects
SDF Observatory
A searchable archive of 178,000 defense-forum posts, with AI-written subject pages and a research assistant that both cite their sources.
BurstSR
An Android camera app that merges a handheld burst of RAW frames into one sharper, cleaner photo on the phone's GPU.
DJI Fly Mod Reconstruction
Rebuilding an obfuscated DJI Fly mod as readable source that compiles back to the original.
HighriseDB
A community-maintained map of 7,500+ skyscrapers in 283 cities, with a forum for every city and building and AI agents that draft sourced edits.
China Industry Visualizer
An atlas of 4,380 source-linked metrics comparing China's industrial scale with other countries, most with year-by-year history.
Experience
- SUMMER 2025
Research Assistant
SUMMER 2025Gong Lab (BU RISE)at Boston University
Built a full-stack research tool for dataset exploration, transcript-video synchronization, annotation, and frame-level search, reducing a teammate’s clip-search workflow from three days to about 20 minutes. Co-authored BabyVLM-V2, accepted to CVPR 2026.
- 2024 — 2025
Research Assistant
2024 — 2025Hayes Labat UC Irvine
Developed parsers and evaluation/visualization tooling for PoseBench3D, a cross-dataset benchmark spanning Human3.6M, GPA, 3DPW, and SURREAL. Standardized MPJPE/PA-MPJPE reports for 18 pose-lifting methods, generating 100+ cross-dataset comparisons; co-authored a CVPR 2026 submission.