formerly Google Research

Developed deep contextual bandit, computer vision and imitation learning methods and deployed onto robotic platforms to empower disabled users to feed themselves. Mentored 6 PhD students on experimental, theoretical and dissemination aspects of their projects.

Built novel latent structure identification algorithms using modern computer vision and natural language processing tools to index, track and identify human trafficking victims among almost a billion multi-media documents. Demonstrated conventional cross-validation in datasets with certain generative properties is biased by upwards of 25%, and proposed the first technique for provably correcting this bias. Successfully deployed as part of a lab spin-off which amplified law enforcement agents' performance to rescue thousands of victims across over 50 agencies.

Trained and deployed deep learning models for forecasting pedestrian, cyclist and vehicle locations. Corrected for trajectory velocity bias in training data to improve model performance on fast-moving actors.

Developed a robotic wheelchair test-bed for improving mobility of quadriplegics and other disabled persons. Mapped indoor environments in ROS with LIDAR sensors and designed closed-loop position control for semi-autonomous navigation.Designed and built a hybrid-electric bicycle trailer, raised pre-seed funding to build and test a minimum variable product, and patented design with the US PTO.

Modeled lithium-ion battery energy production pathways. Integrated results into the Greenhouse Gases, Regulated Emissions, and Energy in Transportation model to determine the total energy cycle of internal combustion, hybrid, and electric vehicles. Concluded vehicle electrification decreases net energy consumption by up to 31% for US electricity mix.