At NVIDIA Research, I research autonomous driving architectures and closed-loop training systems, focusing on interaction-aware planning and decision-making using latent world models and reinforcement learning.
NVIDIA AVG lab →Joined NVIDIA to lead production on Alpamayo — an end-to-end stack unifying perception, reasoning, and control in a single learned model.
See Alpamayo →Researched and built the core product at Spanning Labs, a documentation automation startup. Also contributed to an early-stage robotics startup, working on vision for localization and collision avoidance and RL for multi-bot coordination.
Stepped back from industry to deepen my ML skills and published several open-source projects. Traveled through Nepal, India, Thailand, Indonesia, and Kyrgyzstan, trekking the Himalayas and diving with manta rays.
Check out my portfolio →Zoox was acquired by Amazon in 2020. I used the shift to pivot into ML and reinforcement learning.
Read about the acquisition →Investigated learning-based approaches for rules-of-the-road behavior — stop signs, following, and intersection yielding. Also worked on multi-agent RL for coordinated vehicle decision-making.
Stepped back from my lead role to join Zoox's Research Team full time, focused on the forefront of planning and control.
Integrated machine-learned prediction into dense tree searches for planning. Built an on-vehicle demo for executives — the new method was more adaptable and natural, and influenced how the entire Planner team developed features.
Led a new collision avoidance system (CAS) team, taking the project from a one-file Python prototype to a validated ROS component running on the entire fleet. CAS acts as a backup that verifies the AI's output is collision-free.
Demonstrating success required increasing the fleet's maximum deceleration — achieved by tapping Toyota CAN signals and brake potentiometers. Validation included testing with fake pedestrians on the streets of San Francisco. Issued 10+ patents over my time at Zoox.
Moved from Firmware to the Planner team to develop motion planning algorithms — stopping for jaywalkers and crosswalks, anticipating pedestrian light cycles, and yielding at intersections via geometric collision checks.
Later led the high-speed highway driving effort, coordinating across engineering, QA, and vehicle operators. Resulted in a fully autonomous demo for Bloomberg, where I was the safety driver.
Watch the autonomous demo →Joined Zoox as a Firmware Engineer. Wrote safety-critical code ensuring the AI stayed powered on and the safety driver remained in control.
Work spanned requirements writing, HIL tester programming, and system-level validation.
Took a hiatus from grad school to intern on the Autopilot team at Tesla, joining the core group pushing Autopilot v1 out the door. Traveled frequently to LA to test changes on Elon's 405 commute into SpaceX.
Integrated in-house learned fleet maps into the lane-keeping algorithm, a fallback when cameras lost lane lines.
Returned to UC Davis for my M.S. in Mechanical & Aerospace Engineering. Projects included an autonomous obstacle-avoiding tractor, a Kalman filter for relative position estimation, a CNN-based restaurant classifier, and a satellite design to reboost the Hubble Space Telescope.
Worked as a Manufacturing Engineer at Rolls-Royce Motor Cars in Chichester, England. Based in the leather shop, optimizing manufacturing processes for associates in low-volume settings and developing visual communication standards for presentations and documentation.
Interned at Tesla on the purchasing team during Model S production. Built an internal supplier scorecard to evaluate performance and inform upgrade or phase-out decisions. Also created work instruction videos for Model S manufacturing.
Developed my interest in dynamics and control through undergraduate coursework, then applied it hands-on with Baja SAE — building a single-rider off-road buggy to compete in a 4-hour endurance race.
The first year we over-engineered the vehicle and failed technical inspection. I became captain the following year, simplified the design, fabricated the drivetrain, passed inspection, and competed.
First exposure to programming in a high school CS class. Built a tower defense game and was hooked by the leverage software gives you over a computer.