stephen

hung.

Berkeley EECS ’28. 12-time hackathon winner. I build at the intersection of AI agents, real-time systems, and the protocol-level infra it takes to make autonomous software actually do things in the world.

Stephen Hung at the OpenAI hackathon

Comfortable across the layers — happy at a shader, equally happy at a request handler. 12× hackathon winner. Most recently: pump.fun ($250k, Opal), B@BHacks 1st place (Lapis), OpenAI Codex top 5 (Yolodex, demoed to Sam Altman).

services.
  • Agents
  • Interfaces
  • Infra
  • Automation
  • Systems
Berkeley
OpenAI
NVIDIA
pump.fun
Y Combinator
Cal Hacks
HackHarvard
Sui
Solana
XRPL
ElevenLabs
LiveKit
Letta
Convex
Browser Use
Berkeley
OpenAI
NVIDIA
pump.fun
Y Combinator
Cal Hacks
HackHarvard
Sui
Solana
XRPL
ElevenLabs
LiveKit
Letta
Convex
Browser Use
Stephen at Y Combinator
Stephen at Yosemite
stack.
ReactNext.jsTypeScriptTailwindGSAPThree.jsFastAPIBunConvexExpressClaude APIGeminiOpenAI CodexLiveKitElevenLabsYOLOSAMCLIPmanimCOLMAPSui MoveXRPLSolanax402VercelRailwayDocker
faqs.

EECS at UC Berkeley, class of '28. Started shipping in middle school — a JavaScript Mother's Day card, then VEX robotics captain (PID + odometry for autonomous routines), then FBLA nationals (mobile + computer programming), now hackathons. Valedictorian out of Ayala HS, GPA 4.69. Currently dev at Blueprint (tech-for-nonprofits) and previously SWE intern at OptiGenix (Skydeck Pad-13 alumni).

Tenzin (a personal Claude Code daemon running 24/7 with Discord, Telegram, voice, cron, and multi-session thread isolation), Lapis (AI due diligence + prediction market settled on XRPL), Opal (AI gaming companion you can queue up with), and a few hackathon-stage things still cooking. Always one project shipping, one being rebuilt, one fresh.

Yes — open to summer 2026 internships and post-grad full-time (May 2028). Most interested in agent infra, voice/realtime systems, or anything at the protocol level. Email me with the role and I'll reply within 48h.

If it's a hackathon — pick something that demos in 90 seconds and could keep going for a year. If it's a personal project — pick the part of the stack I haven't shipped before. Either way, the rule is novel tech + a constraint that forces taste. Lapis came from "can crypto-conditions enforce vesting without lawyers," Opal came from "can an agent actually queue up with you," Yolodex came from "can you turn a YouTube clip into a trained YOLO model overnight."

Bun > npm. Vite > CRA. TypeScript strict. FastAPI for the backend, Next.js for the frontend, GSAP for motion. AI work in Python or Bun depending on which one's faster to ship. I optimize for how fast I can rebuild something, not how clever the architecture looks.