• Spearheaded the migration of a critical database storing millions of robotic work requests to AWS DynamoDB, releasing 2,000+ EC2 hosts running Apache Cassandra and saving ~9,000 engineer hours annually • Engineered the data processing infrastructure for a diagnostic dashboard which enables operation teams to root cause 98% of performance gaps on Amazon robotic floors and attracts ~68k monthly visits• Designed a periodic AWS Lambda sweeper for executing essential business… Show more • Spearheaded the migration of a critical database storing millions of robotic work requests to AWS DynamoDB, releasing 2,000+ EC2 hosts running Apache Cassandra and saving ~9,000 engineer hours annually • Engineered the data processing infrastructure for a diagnostic dashboard which enables operation teams to root cause 98% of performance gaps on Amazon robotic floors and attracts ~68k monthly visits• Designed a periodic AWS Lambda sweeper for executing essential business operations on a tier-1 scheduling service, resulting in $45k annual savings and efficiently processing ~1.5 billion work requests per minute Show less

["Athletic Mobile Manipulator System for Robotic Wheelchair Tennis". IEEE 2023]• Led the development of a robotic tennis platform, featuring a racket-wielding robotic arm on a motorized wheelchair, and devised its ball-tracking system, adaptive swing controller, and robust motion planner["Learning Efficient Diverse Communication for Cooperative Heterogeneous Teaming". AAMAS 2022]• Improved message compression by 200x within a multi-agent reinforcement learning model (HetNet)… Show more ["Athletic Mobile Manipulator System for Robotic Wheelchair Tennis". IEEE 2023]• Led the development of a robotic tennis platform, featuring a racket-wielding robotic arm on a motorized wheelchair, and devised its ball-tracking system, adaptive swing controller, and robust motion planner["Learning Efficient Diverse Communication for Cooperative Heterogeneous Teaming". AAMAS 2022]• Improved message compression by 200x within a multi-agent reinforcement learning model (HetNet), leading to more efficient communication between coordinating agents with distinct roles["Fast Lifelong Adaptive Inverse Reinforcement Learning from Demonstrations". CoRL 2022]• Designed the real-time motion controller, simulator, and perception system for a Ping-Pong robotic platform that learns from heterogeneous human demonstrations and isolates user-specific strategies["DROID: Learning from Offline Heterogeneous Demonstrations via Reward-Policy Distillation". CoRL 2023]• Implemented inverse reinforcement learning algorithms to generate high-quality paths for NASA’s Mars Curiosity Rover by distilling the implicit expertise and personal strategies of trained human rover planners Show less

• Developed the real-time mapping, optimization-based path planning, and vision-based opponent-vehicle detection capabilities of a high-speed autonomous electric go-kart for the evGrand Prix national competition (2nd Place 2021) • Enhanced the perception stack of an autonomous 1/10-scale RC car by improving its detection accuracy of road lane markings, traffic cones, and road signs for the IARRC international competition (1st Place 2019)

• Designed hierarchical action primitives, real-time motion controllers with force feedback, and tool-changing routines for a robotic arm (UR10e) to disassemble a motor mount for NASA’s HOME Project• Researched learning from demonstration (LfD) approaches involving image translations to convert human video demonstrations to robot actions to directly learn human strategies through context analysis

• Designed the company website which attracts 500+ monthly visits and is the source of ~70% of new patients• Built a web application to synchronize patient appointments between multiple medical calendars.

• Researched the applications of statistical bootstrapping on dynamic molecular systems to accurately approximate thermal conductivity properties while utilizing fewer computational resources and simulation trials.