About Me

The question I care most about is this: can we make LLMs genuinely human—not just fluent, but able to simulate how feelings, beliefs, and lived experiences shape the decisions people actually make? And if so, can we use that to build society-scale simulations that help us navigate a complex world full of uncertainty?

I came to this through an unusual path. I studied physics, where I was fascinated by chaos—the Lorenz attractor, the butterfly effect, how a small perturbation can cascade into systemic change. After graduating, I spent five years building database kernels and distributed systems (Redis internals at Kingsoft Cloud, infrastructure at Grab serving 44M+ users), where the core challenge was the same problem in engineering form: designing systems robust enough to withstand cascading failure, because a large-scale distributed system is itself a chaotic system.

Then I encountered Nassim Taleb and complexity economics, and the question shifted from infrastructure to society: our world is fragile, black swans arrive without warning, and the decisions that matter most are the ones we can't safely experiment with in the real world. Classical agent-based models like Sugarscape already showed that simple simulations can reproduce real-world distributions. Now, with LLMs and richer data, can we go further?

That question led me to design Sugarscape simulations where 100 LLM agents develop moral dispositions through dialogue alone, and to show that a single cooperative seed can shift an entire population's norms. I bring production infrastructure instinct, physics training, a deep interest in cognitive science (Dehaene, Kahneman), creative practices—science fiction and photography—and a lifelong habit of building communities, from founding a mental health initiative at Grab to growing open-source AI research networks.

Simulating human behavior well requires more than technical skill; it requires genuine attention to how real people connect, decide, and shape each other. I want to contribute to this wave.