An introduction
I’m Felix Yu.
I have always been curious about how things work: how the natural world comes about, and how intelligence arises within it. That curiosity has led me, again and again, to artificial intelligence and the foundations of physics. I have spent much of my life thinking about these questions and pursuing them through projects of my own.
Professionally, I am a startup founder. For the past decade, I have been building a fintech company that makes software for investors. More recently, I have brought agentic AI into that work, designing the systems around models—the tools, instructions, and feedback that help them carry out useful research.
This site is a place to bring my ideas and work across subjects together, both for my own reference and for anyone who shares these interests.
AI and machine learning
I studied applied mathematics at Columbia and Stanford. My interest in AI began during those years and has since grown into more than a decade of technical projects. I have especially enjoyed machine-learning competitions, becoming a Kaggle Competitions Master along the way.
Reinforcement learning has been at the center of my AI work. My projects include RL transfer learning in OpenAI’s Sonic the Hedgehog competition, which took second place in the write-up category; Sim-to-real for Donkey Car Project; and a self-play football agent that earned a silver medal in Kaggle’s Google Research Football competition. I have also implemented and open-sourced many well-known RL algorithms, including the C51 Distributional Bellman Algorithm.
Before the LLM era, much of my deep-learning work centered on computer vision. That included gold-medal work in Kaggle’s Nature Conservancy fisheries competition and Vesuvius Challenge, as well as vision-based self-driving experiments with Donkey Car. My open-source projects have accumulated more than 2,000 GitHub stars.
Lately, I have been drawn to LLM reasoning and agent harness design. I earned a silver medal in the Kaggle AI Mathematical Olympiad, and I built a custom agent harness from scratch for stock research at my company.
Foundations of physics
Physics is another lifelong passion. I am especially interested in the foundations and interpretation of quantum mechanics. I want to examine the ontological and epistemic roles of the wavefunction, Schrödinger dynamics, and Hilbert space, alongside the constraints imposed by Bell’s theorem. Quantum theory underpins the extraordinarily successful Standard Model, but its conceptual foundations still invite debate.
I am developing a picture that brings together Hamilton–Jacobi mechanics and Gibbs ensembles. In it, a particle may be a soliton-like structure sustained by nonlinear dynamics, with the Schrödinger equation describing aspects of its emergent statistical behavior.
The wavefunction would then package statistical information about the particle’s whereabouts together with phase information tied to those underlying dynamics. In this picture, the Born rule is a natural outcome of mapping an ontological probability space into the complex Hilbert-space formulation. This is work in progress, not an established result; I will develop it further in the physics section. Essays are forthcoming.
Elsewhere on this site
I also write short essays and miscellaneous pieces on AI, physics, philosophy, startups, and other subjects. My research manifesto sets out the questions that currently animate my work. If you are thinking about similar problems, or see a possibility for collaboration, I would be glad to exchange ideas by email.