Randomized permutation training for TARFlow · pixel & latent space · Prof. Jiatao Gu

• 3.4× lower FID on CIFAR-10 vs baseline (287 vs 969) with annealed permutation training, matching NLL within 0.5%
• Gap isolated to log|det|, not MSE(z): order diversity improves flow expressiveness without hurting reconstruction
• Current: extending to learned latent spaces (RAE-Flow), joint encoder + flow training toward multimodal world models
• Autoregressive flows · DINOv2 · PyTorch · pixel & latent generative modeling

Non-generative world model for autonomous racing · F1/10th · deployed on hardware

• 37.6% mean lap completion · 50% best seed on Levine track
• Berlin ceiling at 57.7% — competing hypotheses: encoder aliasing vs reward sparsity (open)
• PyTorch · ROS2 · LiDAR · PPO · sim-to-real

PPO-based quadrotor control for autonomous racing · NVIDIA Isaac Lab

• Reward design, observation spaces, reset strategies for stable policy learning
• Iterative training with W&B experiment tracking; diagnosed reward hacking failure mid-run
• PyTorch · NVIDIA Isaac Lab / Isaac Sim · Weights & Biases

Deterministic piano-playing robot arm · 1 kHz control loop

• OSQP-solved contact-mode transitions · hierarchical task planner
• Contact-mode state machine handles fingertip–key dynamics without destroying the action mechanism
• C++ · OSQP · ROS2

Working paper · No-regret learning dynamics for Pairwise Pareto Correlated Equilibria

• Identified squared distance to the no-Pareto-improvement cone as a tractable non-convex potential; proved O(T^−1/2) convergence under a Local Descent condition
• Proposed Bottleneck Regret Matching (BRM): scalar regret matching on the bottleneck objective coordinate
• Showed vanishing swap-dominance regret implies convergence of empirical play to PCE / PPCE
• Game theory · Convex analysis · Regret minimization

PPO deep RL for autonomous hardware reconfiguration · GRASP Lab · Prof. Mark Yim

• Morphology and policy co-evolve in simulation, then transfer to physical modules
• Isaac Gym · PyTorch · URDF · sim-to-real

Quantitative Research & Trading Associate Intern · Equity Derivatives

• RL for hedging variance swaps

Product Consultant · Digital Assets & Tokenization

• Evaluated ClearFi digital-asset product; identified PMF gaps and proposed strategic roadmap directions
• Presented tokenization strategy to product and operations stakeholders

Quantitative Research Analyst Intern · $600B+ AUM

• +60bp information coefficient across 75% of equity factors and sectors
• Valuation-dispersion adjustment to improve factor-model robustness under long-horizon regime shifts

Director of R&D · Penn Blockchain

• $15k+ in individual prizes across ETHGlobal, BUIDL AI Seoul, and other hackathons
• Technical consulting; research on RWAs, DePIN, and tokenization

Trading & Analytics Wintern

• Exposure to execution services, market making, trading algorithms, and market-microstructure analytics
• Proposed data-driven cross-selling strategy for Virtu Execution Services based on client behavior

Software Engineering Intern · GTS

• Network Topology Visualization tool for MDU home-internet service — Neo4j · MongoDB · Kafka · React
• Outage-impact “what-if” simulation: −20% MTTR via predictive equipment-failure propagation

Quantitative Research Intern

• Quantified volume and volatility shocks around FOMC releases across equities, fixed income, and FX

Researcher & Consultant · Wharton

• Menon & Menon: strategic research for a drone/UAV startup
• BTCS: authored whitepaper on Ethereum protocol economics
• Automation Anywhere: GTM strategy and AI positioning

Violin and piano, 14 years. Currently listening: Brahms Symphony No. 2.

Books, articles, papers.

• The Last Question — Isaac Asimov
• Absolutely on Music — Haruki Murakami

Coming soon: a post on the state of robotics in the EU.