I turn math into systems that perceive and move.

Most of my time goes to autonomy: teaching robots to navigate and perceive when the easy signals disappear. That has meant path planning and sensor fusion that lean on inertial and visual data instead of GPS, generative models to simulate situations you can't safely collect data for, and squeezing real time performance out of onboard hardware with CUDA and classic estimation like Kalman filtering.

Before that I lived in GPU performance at Nvidia, writing performance critical C++ for deep learning inference. Dynamic shapes, operator fusion, and quantization to make transformer models run at the speed the hardware actually allows, plus the compiler tooling and test automation to keep accuracy honest across changes.

Earlier I built computer vision pipelines that caught manufacturing defects on a production line, and recommendation systems driven by learned embeddings. The through line started in my EE days at UT Austin: I like understanding a system from the circuit up to the model on top, and reaching for math when a problem looks intractable.