Eleven Seconds
On the Monad testnet, each contract upgrade waits 30 blocks while three sentinel processes attack the new code in simulation and veto it on chain if it breaks a rule. Built on site, entered solo.
Ted Chen · Independent AI Engineer · Taiwan
I build voice AI, LLM agents and tools that check smart-contract upgrades before they run. Claude Code and Codex write most of my code from my designs, and I review, test and ship it.
I started writing algorithmic trading strategies in 2006 and smart contracts in 2016, and have focused on AI systems since 2023.
On the Monad testnet, each contract upgrade waits 30 blocks while three sentinel processes attack the new code in simulation and veto it on chain if it breaks a rule. Built on site, entered solo.
On the Ethereum Sepolia testnet, a proxy upgrade has to wait out a cooldown and get an on-chain clearance. A Chainlink CRE workflow attacks the new implementation in simulation first, then clears or vetoes it. Entered solo.
Each day it opens YES/NO markets on the Monad testnet for a city's high temperature, one per strike, and settles them from airport METAR reports through a Chainlink CRE workflow. A market maker quotes on Kuru order books, and AI agents can trade through a MetaMask Agent Wallet plugin. 149 Foundry tests.
A three.js scene where a Tripo-generated Earth rises over a lunar crater rim rebuilt from real elevation data.
Voice ordering on OpenAI Realtime over WebRTC, shown in a demo store. The server checks every tool call the model makes and recomputes prices. The case study includes a 90-second recording of a full voice order, from greeting to the kitchen display.
A personal prototype that runs a voice AI agent on an ESP32-S3 AMOLED watch, built on xiaozhi-esp32 (MIT) and ESP-Claw (Apache-2.0) with my own Go voice gateway. When answers kept getting cut off, a control run ruled out speaker echo and traced the cause to barge-in with no confirmation window.