hello, my name is
About
I'm a biomedical engineering student at the University of Michigan, minoring in computer science.
My interests lie at the intersection of the body and a machine, with the goal of improving quality of life.
Currently, I'm a systems engineering intern with the Department of the Navy, applying deep learning to medical device signal data, and an exoskeleton research intern at the Locomotor Control Systems Lab.
Previously, I've worked as a summer research intern at the University of Illinois Chicago Rehabilitation Robotics Lab and at UChicago Medicine, where I contributed to projects spanning wearable robotics, prosthetic systems, and biomedical research.
Work
A powered ankle-foot prosthesis taken through 10 human subject trials: TwinCAT operation, Vicon calibration, and synchronized EMG, ECG, and metabolic cost capture. Threaded encoder caps designed in SolidWorks and CNC machined.
A signal-agnostic open source AR platform that pixel-streams arbitrary LSL biosignals into a HoloLens 2; sole developer. Motion-to-photon latency was 120 ms on the headset versus 119 ms on a wired monitor (validated against injected delays up to 1 s), and in a two-participant pilot, AR step-length guidance reduced gait asymmetry by 41% and 53%.
A wrist-worn Parkinson's tracker on a FREE-WILi that runs a one-minute MDS-UPDRS motor check (hand flips, tremor via FFT, alternating taps, voice) with ElevenLabs voice guidance. Results are tagged with time since the last dose and stored in TimescaleDB so clinicians can see medication kick in and wear off, while Fetch.ai agents and iMessage alerts keep caregivers in the loop.
Parametric locking compression plate designed in SolidWorks and validated to ASTM F382 via finite element analysis. Bending FOS 1.73, torsion FOS 2.36. Parallel Simulink model simulates fracture healing load transfer over 12 weeks.
A 3D-printed two-finger gripper driven by surface EMG: MG996R servo, MyoWare 2.0 sensor, and Arduino Uno on a separate supply to stop servo stutter under load. An LDA classifier on five time-domain features separates rest, open and fist at 99.4% accuracy, trained on 9,000 labelled windows.