Hey, I'm Antonio. I'm studying EE & CS at Stanford and am interested in alternative compute, embodied ml, hci, and world models. Prev 3x IOI/IMO training camps, 1x natl computing champion.
Sequential Decision-Making in DNA
DNA circuits can compute, but most implementations are static logic gates. This project explores sequential decision-making: circuits that observe biomarkers, maintain internal state across multiple measurements, and trigger outputs only after accumulating sufficient evidence. The pipeline compiles POMDP policies into finite-state controllers, maps controller nodes to strand displacement cascades, then uses RL to assign nucleotide sequences that optimizeboth thermodynamic stability and reaction kinetics. The motivation is enabling more nuanced molecular computation for possible future use cases for in-situ diagnostics and therapeutics.
in progress. [milestone]
Real-Time 2D Game Generation Without Engines
I'm interested in: Exploring whether 2D games can be generated in real-time without traditional engines. Our approach trains action-conditioned diffusion models on gameplay from PPO agents, then aggressively compresses them for consumer hardware through smaller latent spaces, single-step sampling, and simpler attention. Testing if 2D visual simplicity admits efficient architectures that preserve game dynamics under tight latency budgets. Comparing multiple methods to understand quality-compute tradeoffs.
in progress. pre-print expected end of december.
