I'm focused on self-supervised learning and world models.
- Opportunistic Target Selection - CAp 2026 accepted paper on query-efficient black-box adversarial attacks.
- DashVMC - Ongoing research on real-time discrete world model control in Geometry Dash with FSQ tokenization, action-conditioned transformer dynamics, and imagined-rollout policy training.
- Hack the World(s) - Finalist at a Yann LeCun-sponsored JEPA/world model hackathon; built a self-supervised EEG system with Riemannian analysis of the learned representations.
- Rose - Open-source League of Legends customization tool with 25K+ daily active users.