Reproducible Vulkan optimization evidence for Llama 2 7B Q4_0. The retained profile improved decode from 75.23 → 85.71 tokens/s with flash attention off.

Repository · Results

02 / Coup RL bot Deep RL strategies for the card game Coup, with bluffing, challenges, counterplay, and a playable browser demo. Repository · Play the bot

03 / Heston regime lab An interactive lab for calibrating the Heston model to SPX options, detecting regime shifts with a hidden Markov model, and visualizing how parameters change across regimes. Repository

Inference Systems · GPU Kernels · Reinforcement Learning · Agent Evaluation · Systems Optimization · Data Visualization · C++ / ROCm · Python · Research Design

I’m an undergraduate at UCLA studying Applied Mathematics and Statistics / Data Science. Mathematics first drew me in; building projects made the ideas tangible. That combination keeps pulling me toward machine learning and optimization.

Away from the keyboard: weightlifting, soccer, and piano.