experience

University of California, Los Angeles

• Dissertation: Learning How and Why: Causal Learning and Explanation from Physical, Interactive, and Communicative Environments

University of California, Los Angeles

• Thesis: Learning Complex Functional Manipulations by Human Demonstration and Fluent Discovery

University of Dayton

• Thesis: High-Performance Declarative Memory through MapReduce

General Motors · Remote

• Develop end-to-end driving models trained with imitation and self-play reinforcement learning.
• Joined GM through its acquisition of Cruise.

Cruise · Remote

• Co-developed DriveGPT, scaling autoregressive behavior models for driving to 1.4B parameters and 120M training segments (~50x more data than prior published models); demonstrated data and model scaling laws for driving (ICML 2025).
• Applied imitation and reinforcement learning to trajectory generation for motion planning.
• Distilled large behavior models into compact models for on-vehicle deployment.
• Co-inventor on a patent application for autoregressive models for autonomous agents (US 2026/0131822 A1).

Cruise · Remote

• Worked at the intersection of machine learning and motion planning on the Behavior Planning team, starting with maneuver planning.
• Applied imitation learning to trajectory generation for behavior planning; used PyTorch and ROS to deploy safer, more capable ML models to vehicles driving on public roads.

Center for AI and Robot Autonomy (CARA)

• Defined updated mission and vision statements, sought alternative funding, and maintained 501(c)(3) status.

Computer Science Department, UCLA · Los Angeles, CA

• Causal learning: built OpenLock, an open-source virtual escape room in which every room shares an abstract causal structure while surface features change. Ran human-subject experiments showing people discover and transfer that structure, and developed a hierarchical Bayesian model that matched human transfer while seven model-free RL algorithms failed (AAAI 2020 oral; CogSci 2018 oral; CogSci 2019).
• Imitation learning: taught a robot to open medicine bottles from human demonstrations by integrating force and pose, captured with a custom tactile glove and planned with an embodied haptic model and a stochastic grammar (IROS 2017, two papers).
• Explainable AI: extended the bottle-opening robot to explain its own behavior; showed that the explanations that best foster human trust differ from the components that drive task performance (Science Robotics 2019).
• Also contributed to abstract causal reasoning (CVPR 2021) and in situ human-robot value alignment (Science Robotics 2022).

Santa Monica College

• Taught one or two 45-student classes per quarter: led lectures, held office hours, and created course materials.
• Instructed 25 courses: Internet Programming (HTML, CSS, JavaScript, MySQL, and PHP), Intro to C, and Intro to C++. See teaching.

Center for AI and Robot Autonomy (CARA)

• Worked on a representation learning project coupling vision and language (see research).

Computer Science Department, UCLA

• Introduction to C and Introduction to C++: led discussion sections (~50 students) and held office hours.

Electrical & Computer Engineering Department, University of Dayton

• Electronic Devices Lab: aided in lab sessions (~40 students).

Garmin International

• Automated the testing process for small-craft airplane ACARS systems that send timed status messages to ground stations.
• Reduced testing time by 40% and saved hundreds of vendor certification testing hours by optimizing simulation timing protocols while adhering to FAA safety standards.

School of Engineering, University of Dayton

• Tutored first-year engineering students in calculus, chemistry, and physics; led a team of 8 tutors overseeing 40 students.

Cristo Rey Kansas City

• Taught four junior and senior classes at a prep school focused on college placement for underrepresented groups.