tommy
! decision log · 2026

Tommy (Haotian Zeng) · press the hat ☻

I SAID
NO.

¥800K/yr offer — declined. NTU place — declined. Plan B — deleted.

At 22, three months out of school, I went all-in on a startup and a one-way move to Silicon Valley.

Most people use AI to write code faster. I'm building toward the layer where AI builds the whole thing.

// what i gave up

- ¥800,000 / year   // signed offers, highest of several
- NTU, Singapore   // a place I'd worked years for
- a salary, a visa, a plan B
- the version of me my family could explain at dinner
+ one bet: build the thing I can't stop thinking about

People keep asking if I'm scared. Honest answer: the offer scared me more. Comfort at 22 felt like a slow way to never find out what I'm capable of.

// what i'm building

I'm co-founder at Mindfold (mindfold.ai) — growth & product. We build Trellis: the intelligence layer for AI-native engineering. It turns shipping AI software from a black box into an observe → discover → improve loop.

trellis ★ 11K+ · 630+ forks · last push recently fetched live ↗

That number moves while you're reading this. That's the point — I ship things that keep moving.

// the receipts

I don't plan things to death. I ship, then the world tells me if I was right. The pattern, newest first:

Co-founded Mindfold (growth & product) & help maintain Trellis' core — took the agent harness open-source to 11K+ stars in months. [mindfold]

Principal engineer on ~half of Meshy's Creative Lab: turned language & images into manufacturable 3D (boolean ops → mesh repair → printable supports). [meshy · where I sharpened the knife]

Built the world's first flying AI "company pet"; my demo helped close a ¥12M seed and 10×'d the valuation past ¥100M in a month. [skyris ai]

Shipped a cross-platform AR + web multi-agent workflow tool for one-person companies — solo, on MCP. [shitu · founder]

Built clinical sleep-medicine AI for Zhongshan Hospital (Fudan); cut annotation effort 80%, contributed to a pending patent. [clinical ai]

Published glaucoma detection at 94.2% accuracy (ViT + VGG + RF), ICIMR 2024. paper →

Co-founded Accele: neural search at 1M+ daily queries; ¥500K angel; beat Google Custom Search on relevance. [co-founder]

Won across 4 tracks at Adventure X: multimodal medical-consult AI on Rokid AR glasses & Meta Quest 3.

Led 8 people to ship EAPTalk, a WeChat English-learning app; 5,000+ organic installs.

see the pattern? none of these were the safe choice either.

// the dream

I'm building toward an OASIS.

Yeah — the one from Ready Player One. A world worth living inside. Mindfold is the company, Trellis is the product, but OASIS is the thing underneath — the reason I turned down the money. Generative 3D, agents that build, worlds that render themselves. I'm still putting words around exactly what it is. If I could explain it in one sentence, someone would've already built it.

ask me in person and I'll tell you the version I actually believe. _

// find me

H2 2026 I land in Silicon Valley with one bet. If you're a founder or you build near the edge of AI — generative 3D, agents, AI-native tooling — tell me what you're building and I'll tell you what I see.

generative 3Dmulti-agent systemsthe leap itself

the door's open. _