St. Louis · Chicago
I build at the intersection of rigorous math, game theory, and machine learning.
Now
Building a compliance-filing tracker for SVRN and a poker GTO solver on the side.
Selected Work
A masked diffusion language model that maps text to emoji by denoising a whole sequence in parallel, instead of generating one token at a time left-to-right. The wager: diffusion's iterative refinement is closer to how people actually reach for meaning than autoregression, and it turns extra inference-time compute directly into better answers.
82% semantic pass@7 vs ~59% for the best frontier model
Working towards a full poker solver through iteratively more complex game models, starting with Kuhn poker and moving up through Leduc to full Texas Hold'em.
169 hands · 15k iterations · scored by exploitability
A benchmark that grades a model's poker decisions against the solver's equilibrium. Three difficulty tiers, with exploitability as the single primary metric.
3 tiers · exploitability-first scoring
A working computer built from simple logic gates— ALU and memory wired by hand. The point was to leave no layer of abstraction unopened through a deep dive into digital design.
Built from discrete logic, no microcontroller
About
Sophomore at WashU studying CS, math, and statistics. I like problems where structure hides underneath apparent randomness — equilibria, distributions, optimization landscapes.
Most of what's below is in motion right now. I'd rather show the thinking than wait for a finish line.