Howdy :)
I am currently finishing my BS and MS in computer science at Johns Hopkins University, under the guidance of Prof. Peter Kazanzides in the JHU Laboratory for Computational Sensing and Robotics, and Prof. Anton Dahbura in the JHU Sports Analytics & Research Group.
My current research lies at the intersection of computer vision and robotic systems. I’m particularly interested in designing robust multimodal perception and motion planning frameworks for autonomous robots. For more details on my past and ongoing research, check out my CV.
Apart from research, I've also led several SWE endeavors. Currently, I'm overseeing the tech development of a student platform at Harvard Law School. Stay tuned - something big is on its way!
If I’m not doing any of the above, you can probably find me skiing, surfing, or taking a Cessna / Piper up for a quick sunset cruise...
Education
Johns Hopkins University
[Baltimore, MD]B.S. & M.S. in Computer Science, 2022 - 2026; GPA: 3.95 / 4.0
Highlighted coursework: Probability & Statistics, Linear Algebra & Diff Equation, Data Structures, Discrete Math for CS, Computer Vision, Algorithms, Augmented Reality, Artificial Intelligence
Thayer Academy
[Braintree, MA]High School, 2018 - 2022
Highlighted coursework: Calculus I & II, Multivariable Calculus, Java Programming
High school days: Check out this ancient version of my website...
Projects Overview
Dissection Automation, dVRK (daVinci Research Kit)
- Automation architecture: closed-loop motion planning (Action Transformer Model), powered by a multimodal perception stack (Foundation Stereo, SAM3, and a custom contact-detection module).
- Training visuomotor policies from expert surgeon demonstrations; structuring and contributing the training dataset to NVIDIA Open-H-Embodiment to support data sharing across the medical robotics community.
Vision-Based Aerial Ropeway Monitoring System
- Pipeline architecture: YOLO-based multi-person tracking → per-person pose estimation → intermediate layers for skeletal keypoint stabilization and temporal consistency checks → a custom ResNet-based action classifier trained on real-world incidents to detect hazardous loading/unloading events
- Achieved 96.88% classification accuracy in preliminary experiments; demonstrated reliable performance when generalizing to previously unseen scenarios and different viewing angles.
SurgSync: Time-Synchronized Multi-modal Data Collection Framework for Surgical Robotics
- Motivation: addressing the dVRK community’s lack of a reliable, time-consistent data pipeline. Prior methods produced severe temporal inconsistencies when recording multiple sets of kinematic / visual streams. Moreover, a unified post-processing toolkit for preparing dVRK training/validation data was non-existent.
- SurgSync: achieves a stable 10 – 15 Hz recording frequency across 8 ROS topics (each publishing at different frequencies); enables reproducible, high-fidelity dataset collection for visuomotor policy training and validation; has been deployed across labs in the JHU LCSR.
BatVision: Single-View Computer Vision Pipeline for Equipment Dimension Modeling and Quality Control
- Pipeline architecture: precise foreground extraction → custom geometric modeling to extract dimensions → post-hoc GAM-based calibration (replacing checkerboard pre-calibration to accommodate dynamic deployment scenarios)
- Results: sub-millimeter accuracy (MAE 0.148 mm); demonstrated calibration training transferability to unseen equipment.
- Project sponsored by the Baltimore Orioles (MLB); pipeline currently integrated into the Orioles’ analytics stack; received multiple media coverages.
Feature @ CBS News | JHU Hub | Provisional Patent Disclosure