Interests

Some of the domains I spend the most time thinking about and building in.

Research & Papers

ML · Evaluation

I like framing questions precisely, building datasets, and designing evaluation harnesses. I've worked on visual commonsense reasoning with Prof. Ernest Davis, focusing on “visibility” and how models represent what can or cannot be seen.

  • Dataset and annotation design
  • Evaluation scripts across multiple models
  • Error analysis and qualitative examples

AI x Crypto & Agents

Agents · Signals

I work on agentic systems that interact with crypto markets: signal engines, dashboards, and workflows where models, data, and humans all have to line up. I care about transparency, traceability, and hard guarantees more than hype.

  • On-chain/off-chain signal design
  • Agent workflows and safety guards
  • Dashboards that surface what matters

Robotics & Embodied AI

Diffusion · OOD

Currently exploring OOD detection for robot control policies, specifically using score-based methods to detect when a diffusion policy is operating outside its training distribution.

  • Diffusion Policy replication on Push-T
  • TAP-Score for (obs, action-chunk) rating
  • Robustness testing under visual perturbations

Other interests. Competitive gaming (usually top ~2,000 globally in Clash Royale, Grand Champion III in Rocket League, peaked at rank 598 in Marvel Rivals), the gym, and sports like football, rugby, badminton, table tennis, and fives.