I build agent teams that ship production software, and the gates that make AI-written code trustworthy.
agent-loop-v2.mp4
A 7-language product with 1,700+ registered users, built and run alone · median 1.9 h from bug report to resolved fix · 0 of 231 merged PRs reverted · a client's 15 prototype screens made real in 6 days
Read the case study → · Source of the write-up
LinkedIn · Email · Case study · agentic-delivery-kit
Harness bugs I hit while running many concurrent coding agents, reproduced, fixed and sent upstream (why):
Also reported: worktree clones copying .env secrets · disk exhaustion from per-worktree build caches
- Agentic Engineering: Running parallel coding agents with separate author, reviewer, and verifier roles (Oh My Pi, Codex, Claude Code), plus measured code-intelligence context (CodeGraph, Understand Anything)
- LLM products: grounded, schema-validated LLM features with regression evals, cost guardrails and privacy-safe tracing (TriAstra, plus a ChatGPT app served from a Rust MCP server)
- Payments & Web3 backends: subscriptions and refunds, ZK identity, trading automation, token platforms
- Measured trust: evals, typed contracts, and gates that catch errors before runtime
- 🥇 DAO Global Hackathon 2023 (1st Place) – Aragon Most Creative Use of Aragon OSx | GitHub
- 🥇 Aptos Seoul Hack 2023 (1st Place) – Gaming Track | GitHub
- 🥇 Coinbase Hackathon: Buidl The Future (1st Place) – XDC Open Ideas Track | GitHub
- 🥇 Klaymakers22 Global Hackathon (1st Place) – NFT & Metaverse Track | GitHub
- 🥈 BNB Chain Innovation Hackathon Seoul (2nd Place) | GitHub
- 🎖️ ETH Global TOKYO – Polygon Pool Prize & Scroll (Just Deploy) | GitHub
- 📄 Report evaluation device and operating method thereof (inventor, Hyperconnect) | US12137302B2 · KR102503049B1
- 🎓 B.S. in Computer Engineering, Inha University, Incheon, South Korea (Mar 2014 – Aug 2021)
✈️ Exchange Program in Computer Engineering, TU Chemnitz, Germany (Oct 2020 – Feb 2021)




