The places I've been, and what they changed.
2019 — PRESENT · Beijing → Shanghai → Los Angeles
Studied economics for its structure — the way systems could be designed, tested, broken, rebuilt. Found that finance's endpoint was valuation, not creation. Wanted to build something, not just measure it.
Built DCF models. Learned to value companies. Realized I wanted to build them, not value them.
First encounter with generative AI as a building material — a tool that could generate possibilities you didn't know to look for. Started asking the question I haven't stopped asking: what problems is this actually here to solve for people?
Wanted to build something, not just measure it.
M.S. Quantitative Economics. Moved to Los Angeles to be somewhere new — different people, different ways of thinking, room to see the world more clearly. Learning systems design. Reading. Building.
Starting from zero, the idea grew clearer through constant writing and practice. Shared the thinking through documents and a demo — and attracted five interdisciplinary PhD collaborators who joined because they believed in what this could become. Together, built a complete, working product.
What problems is this actually here to solve for people?
Three things currently being made.
Writing is the only thinking you can watch yourself do.
Journaling tools record, but don't respond. Writing is an act of self-expression and thinking — it needs a mirror, not a filing cabinet.
A multi-agent writing companion where AI doesn't write for you — it writes alongside you. Different agents respond to different emotional textures in your text: some grounded in psychology, some in literature, some stranger than that. Real-time, contextual, built around the idea that expression deserves to be met.
100+ beta users. 50% of active writers reported meaningfully higher engagement. 5 PhD collaborators recruited through conviction, accumulated product documentation, and a credible demo.
The hardest part of 0-to-1 isn't the build. It's making people believe in something that doesn't exist yet — and the tool for that is language.
Most portfolios are containers. I wanted mine to have a behavior.
A personal website in the age of AI feels like a new kind of thing — not a resume, not a social profile, but a space that actually behaves like you. I wanted something beautiful, strange, and genuinely mine.
Building this is how I'm learning frontend development, WebGL, and interaction design from first principles. But more than that — I think making a website that reflects how you think and what you find beautiful is one of the more interesting ways to exist on the internet right now.
The system you build determines more than any single effort inside it.
An orchestrated agent framework for personal workflows: capturing inspiration, processing reading notes, structuring thinking through AI dialogue. Not a single tool — a system of agents with clear handoff logic.
I believe in the power of systems — a well-designed architecture creates the conditions for things to grow naturally within it. Not a collection of scattered components, but something with real coherence and life.
Questions I'm sitting with. Capabilities I'm building. Books I'm reading.
How should agents be orchestrated — not just what they do, but how they hand off context to each other?
What makes AI interaction feel like companionship rather than utility?
Where is the boundary between a tool augmenting human thought and replacing it?
What does a personal infrastructure for learning actually look like — one that compounds over time rather than just accumulates?
Emergent behavior, evolution, the beauty of complex systems. Written before AI became a trend — which makes it feel more honest about what complexity actually is.
Human cognition and decision-making. Still reading. The question underneath it: how do people actually think, and what does that mean for how AI should respond?
Memory as reconstruction. Unreliable, selective, quietly beautiful. The question of whether memory is trustworthy feels important — personally and as a design problem.
What I use to build.
I build with AI assistance — Claude Code for development, Figma for design, Python for experimentation. Currently exploring OpenClaw for agent orchestration. Understanding how something works changes how you use it.