• Developing a VR-enabled teleoperation system that reconstructs live 3D environments, allowing users to interact with robots through intuitive point-and-command controls

• Implemented real-time self-collision detection using MoveIt, preventing unsafe joint configurations and improving system robustness during teleoperation

• Collaborated in a small, fast-paced engineering team to prototype and iterate on novel human-robot interaction paradigms

• Developed the perception stack architecture to be sensor-agnostic and modular, ensuring system robustness and long-term scalability

• Implemented a RANSAC-based ground filter, added PCA-derived linearity metrics, introduced new shape-filtering heuristics, and built close-cluster tracking logic to reduce false positives in large but sparse point-cloud segments, improving overall detection stability by >75%

• Improved maintainability and reliability by modularizing code structure and standardizing node interfaces across the perception pipeline

• Designed and implemented Analog-to-CAN and Dashboard software for Formula SAE electric racecar

• Enhanced analog-to-CAN code by increasing modularity, implementing sensor hot-swap capability, and eliminating over 50% of redundant code

• Resolved persistent text display issues and enhanced dashboard visibility for improved user experience

• Leveraged expertise in electronics to support the Electrical team and expedite SPI transmission analysis

• Creator and developer of Sanctuary AR, a full-stack application that maps desktop and mobile tabs and files into rooms, enabling innovative spatial interactions with data via Snapchat Spectacles AR goggles

• Secured full funding from Snapchat to develop the prototype and MVP

• Developed a custom room detection pipeline to monitor and track room locations in real time

• Designed an intuitive user interface, empowering users to detect and label rooms in locations that were not previously assigned

• Developing a bimanual robotic manipulation system in the UW Robotics and State Estimation Lab, focusing on robust real-world teleoperation and data-driven learning.

• Collaborated closely with PhD researcher to improve teleoperation smoothness, reducing control discontinuities and increasing task stability during manipulation.

• Designed and executed targeted data collection pipelines to diversify the training dataset for bimanual valve manipulation tasks.

• Identified and validated the importance of starting end-effector positions below the valve, enabling the learning algorithm to better avoid collisions and significantly improve generalization across manipulation scenarios.

• Contributed insights bridging robot control, data collection strategy, and learning performance, strengthening system robustnes