3D Variational Autoencoder for Neuromuscular Diagnostics
Designing a novel 3D VAE architecture that reconstructs high-density EMG signals from electrical stimulation amplitude to classify Achilles tendinopathy injuries. Building placement-agnostic ML models that abstract electrode positioning—one of the messiest real-world variables in EMG classification—by learning latent representations of muscle behavior under controlled stimulation. Training on 64-electrode HD-sEMG data collected at 7500 Hz across 30+ stimulation intensities per patient.
Python • PyTorch • Signal Processing • Medical Imaging • VAE Architecture
Cable-Driven Hand Exoskeleton for Stroke Rehabilitation
Leading hardware development of a cable-driven hand exoskeleton designed for stroke patients with limited grasping capability. Prototyping an ergonomic single-finger mechanism (index finger, 1 DOF) that assists extension through tendon-like cable actuation, targeting 5-10N force output while maintaining compliance for safe human-robot interaction. Integrating motion capture with IMU sensors to detect user intent and enable closed-loop control. Designing for FDM 3D printing manufacturability with modular expansion to full-hand system.
CAD • FDM 3D Printing • Cable Actuation • Motion Capture • Embedded Systems
Two-Model EMG Classification for Upper/Lower Limb Prosthetics
Building a placement-agnostic prosthetic control system using a two-model-in-series architecture: Model 1 (autoencoder/VAE) learns a latent "placement parameter" from raw EMG to abstract electrode positioning, feeding into Model 2 (CNN/SVM) for gesture classification. Training on open datasets (GRABMyo, EMG1K) with preprocessing pipeline (20-450 Hz bandpass, RMS envelope extraction). Developing real-time control interface with majority voting over sliding windows to smooth predictions for robust prosthetic actuation.
Python • Machine Learning • Signal Processing • Real-time Systems • Prosthetic Control
Brain-Computer Interface for Generative Music
Contributing to a novel BCI system that translates multidimensional cognitive states into real-time generative music as a focus companion. Exploring how neural signals can be decoded to create adaptive, responsive audio environments that reflect and enhance cognitive flow states. Part of Penn Neurotech's spring 2026 project team investigating the intersection of neuroscience, machine learning, and creative expression.
BCI • Signal Processing • Generative Models • Audio Synthesis
Computational Art & Visualization
Experimenting with algorithmic art generation, data visualization, and creative coding. Recent work includes EMG heatmap animations visualizing muscle activation patterns, generative design explorations in Processing/p5.js, and computational approaches to transforming biological signals into visual art. Exploring how technical tools can become mediums for creative expression.
Python • Processing • p5.js • Data Visualization • Generative Design
Microneedle-Integrated Hydrogel Platform
Engineered a minimally invasive biomarker sampling platform combining microneedles with optimized hydrogel formulations. Developed Python-based analysis pipelines for processing 75+ immunoassay samples, performing end-to-end data processing to quantify biomarker concentrations. Systematically optimized hydrogel composition through design-of-experiments to maximize interstitial fluid collection efficiency for clinical trials.
Biomaterials • Python • Design of Experiments • Clinical Research