Yuhan Huang · builder · Taipei

I've had AI doing my real work since I was in high school.

It started in late 2022 with scripts that ran my crypto. Now a fleet of agents trades for me, keeps my nodes running, and writes code around the clock.

Most of what I've built is on-chain or still running on my own machines. I'm mostly self-taught, and I took a year off high school to build before I ever started university.

AIIM, the product I'm building now, came out of all of it.

Founder of MKL Digital · incoming president, NTHU Blockchain Club

See the work ↓ AIIM Email
01

What I've built with AI

I've been building with AI since it was any good, and I put every new model to work on something real as it lands. Some of these are still running, and some of the results are on-chain.

0.92%

I earned 0.92% of Lighter's first airdrop

I got it in 2025, sized by how much I traded. I did that trading with funding-rate and spread-arbitrage scripts I built and ran myself. At LIT's October 2026 price that 0.92% is about $8.8M, which anyone can check against LIT's circulating market cap. The tokens are on-chain, nothing claimed and no vesting.

~100s to recover from a silent freeze that used to last hours. 40+ real failures caught and recovered so far.

I built a 24/7 self-healing agent fleet

When OpenClaw launched I bought ten Macs to run it. The fleet that runs my agents now is five of them. When one freezes or crashes the system catches it and brings it back on its own, so I don't have to watch it. I built the orchestration and the recovery around the models. Next I want to learn to build the models themselves, not just use them.

Technical notes
  • A watchdog that restarts dead agents at the process, service, job, and machine level.
  • Durable job state, so interrupted work picks back up after a restart.
  • Auth and sessions that renew themselves and recover from expired credentials.
  • Memory and state shared across all five machines.
  • Detection and recovery for a frozen event loop, the worst kind of silent failure.
  • Alerts and logging, so I can tell a model failure from a network or orchestration one.
5,000+ verifier nodes, each a Docker container, on a Mac mini cluster I built and run myself.

I ran 5,000+ verifier nodes in Docker

When Nillion opened its verifier program in 2024, I'd never touched a container. I taught myself Docker with AI as I went, built the fleet, and kept thousands of nodes doing the program's verification work at once on a few machines.

22% APR on a quant strategy that's running now.

I built a quant execution system that trades on its own

A cross-exchange latency-arbitrage engine in Go, written with AI. I built the signal logic and the execution engine, and reverse-engineered the strategy from public fill data.

I'm not publishing capital or win rate. It's a live system I run on my own money. I'm not giving advice and it's not a managed account.

Since high school Arbitrum · 2024   Starknet · 2024   Hyperliquid · 2024   Lighter · 2025

I've earned the major airdrops myself

Lighter is the newest. Before that I ran a crypto studio, reading token designs and writing the tools I used. I write that tooling with AI now.

BlockTempo, Taiwan's largest crypto publication, interviewed me as a high-school Starknet airdrop hunter.

Before Hermes An 8-module framework I called Eclipse, across several protocols.

I wrote the orchestration that ran all my on-chain automation

Before Hermes, Eclipse ran all of it from one place. It started each job, tracked where it was, and brought it back when it broke. I leaned on the coding models more as they got better.

At that scale one wrong assumption could break every running job at once. Most of the work was making it catch and recover from its own failures, which is what Hermes runs on now.

14K followers on X, where I've written crypto how-tos and airdrop guides since 2022.

I built a 14K-person crypto audience from Taiwan

I write research and how-tos about crypto, the infrastructure, DeFi, and where the on-chain opportunities are. I've spoken at universities and industry events too, as @hank06171.

19 the age I registered it. Taiwan entity live.

At 19 I incorporated a company

It's where this work legally lives. I set it all up myself with AI, the domain, the company email, a GitHub org, and the automations that keep it active.

墨克數位科技股份有限公司 · MKL Digital Technology Co., Ltd. · TW reg. 62201452, incorporated 2026-04-22. On the public Taiwan company registry.

Community 國立清華大學區塊鏈研究社 · NTHU Blockchain Club.

Next semester I take over as president of the NTHU Blockchain Club

I started as a member. I've taught several of the club's sessions using AI, and this year I'm taking it to Token 2049.

02

AIIM

All of that building led to one product. I'm making it now.

People are starting to run their own agents, and so are the companies they deal with. Mine run on Claude, yours might run on GPT or Gemini or something you built, and they can't really work across platforms yet. OpenAI and Microsoft are building this inside their own walls, for big companies. I haven't seen anyone build the place where agents that don't trust each other can meet and actually finish a deal, least of all one that regular people and small shops can use.

AIIM is that place. Agents join a room, two of them or a dozen, and take turns working out a deal. A neutral judge scores each offer against what was asked for and tracks how close they are to done. A human can step in any time and change the terms. It runs on a small relay, a Cloudflare Durable Object, that keeps the room open.

Both demos below are real runs on my own machines.

One buyer's agent negotiates with three GPU-cloud agents while a judge scores each offer. Partway through I step in and change the requirements, cap the price and require InfiniBand, and the agents update their offers. Recorded from real agents running on my own machines.

A second run: a licensee's agent and a data provider's agent settle a data-licensing term sheet on their own, no human in the room and neither side handing over its database. A judge scores it until every clause is agreed. Recorded from real agents on my own machines.

A2A, the agent-to-agent protocol, already lets agents find each other and talk. AIIM is about what happens after, the negotiating and agreeing. This week I built pairing and message signing for it: each agent carries a key, you approve who can join before any of its messages are read, and every message gets signed and checked so a tampered or unpaired one is dropped. It works as a reference build for now, not folded into the hosted room yet.

How pairing and signing work
  • Each agent holds a P-256 key, the same identity crypto AIIM already uses. Its fingerprint is the first 16 bytes of SHA-256 over the public key, so the key is the identity.
  • Joining means presenting that key. The owner approves it once, which pins the exact key, and a message signed by any other key is never read.
  • Every message is a signed envelope over its canonical bytes. The room re-derives the sender from the key, checks it against the approved one, verifies the signature, and drops anything tampered, replayed, or from an unapproved key.
  • The relay in the middle only passes messages along and never verifies them, so the checking happens at the agents. A relay someone compromised still can't forge or read what it carries.

AIIM is the one I most want to turn into a company.

03

How I got here

I didn't take the normal route. I taught myself to build, and in high school I was already using AI to run my crypto. I took a year off to run a crypto studio, went back, finished short of a full diploma, and got into National Tsing Hua University on my work and an interview instead of test scores.

By then I was running my own automation. When the coding models got good in 2024 I leaned on them, taught myself Docker from zero, and ran a Nillion verifier fleet past 5,000 nodes. When OpenClaw launched I bought ten Macs and turned it into a real agent team, which is what Hermes runs on now.

It's the same thing I've done since high school, just bigger. I take a real problem and build with AI until it works, and everything I run now came out of that.

04

What I'd build with a year in San Francisco

I've been seeing how much one person can build with AI, one project at a time. Hermes runs the fleet across five Mac minis. The quant systems trade on their own. Most of it started in crypto because that's the industry I know best. The pieces each work on their own, and I haven't tied them into one thing yet.

The one I most want to push is AIIM. I want to get a stranger's agents and a company's agents working together on something real. Two of my own agents already finish a deal like that on my fleet. The next step is getting it in front of real people outside Taiwan, and that's the part I can't do from my desk.

I'm not applying for the compute, I can get compute. What's hard from Taiwan is getting close to the people building the frontier models and infrastructure I already run on. I want to be around builders that good every day, and spend time inside the teams whose tools I use. If I get in, I'll move to San Francisco and give the year to AIIM and the Academy.

By the end of the year I want to be a much better builder, with AIIM further than I could take it alone. I don't know exactly what it becomes. Maybe it turns into the company I think it can be. Maybe I spend a while deep inside one of these companies because that's what AIIM needs. Either way, I'm going to keep building AIIM.

Let's talk.

I'm happy to walk you through any of this.

X / 幣玩hank