Computer vision, robotics, and the neuroscience of how systems sense and act.
I build computer vision and robotics software, and I'm headed toward intelligent biomedical hardware: prosthetics and assistive robotics. Cognitive science and computer science at Columbia (Barnard College). Looking for a summer 2027 robotics or health-tech internship.
Since I was a kid I've been obsessed with how things work, whether that's a shoulder joint or a circuit board. At some point it stopped being a hobby and became the way I see everything. When I hit a problem, I see the system it lives in: what feeds it, what it's connected to, what breaks downstream.
The useful part is that systems repeat. A nervous system and an embedded device are solving the same problem: sense the world, move a signal, decide, act, and correct when the result is off. I learned that loop in my own body as a climber before I had words for it, then found it again in neuroscience, and again in sensors and motor control. That's why I study cognitive science and electrical engineering at the same time, and why I'm headed toward intelligent biomedical hardware, like prosthetics and assistive robotics, where the nervous system and the machine have to work as one system.
Columbia University
Barnard College
BA Cognitive Science (Embodied Cognition), Minor in CS. Expected Dec 2027. Dean's List, Spring 2026.
CU Boulder
Online, concurrent · planned completion May 2028
Graduate coursework toward the MS in Electrical & Computer Engineering, Embedded Sensors and Motors pathway.
Robotics Studio, Advanced Neuroscience Lab, Machine Intelligence, Natural Language Processing (PyTorch), Biological Image Computing (graduate level), UI Design, Data Structures
Graduate: Sensors and Sensor Circuit Design, Supervised Learning
Gaze-guided grasp prediction for prosthetic hands. Multi-grip prosthetic hands make the user switch grips by hand. Using MeganePro (eye tracking plus forearm sEMG from 45 people, 15 with below-elbow amputations), I'm testing whether where someone looks on an object predicts the grip they'll use, and how early before the reach, so the hand could pre-shape itself. The predicted grip drives a simulated hand in MuJoCo. Public de-identified data, no new collection.
Preliminary analysis, MeganePro subject S021: trial segmentation, gaze timing around movement onset, per-object grip classifiers from gaze with permutation tests, object-relative gaze from detected boxes, and a table-layout confound check. Notebook (PDF) →
Machine Intelligence, Columbia. 133 build-log entries, three failure classes, a retrieval prototype.
Philosophy of attention, Columbia. An architectural constraint on selection-for-action.
Biological Image Computing (graduate level), Columbia. Cellpose pipeline with an honest failure analysis.
10th at the North American Championships and 14th at the Pan American Championships in bouldering, 2026. Competed at national team trials for a spot on the World Cup team.
3× USCSA All-American: three top-10 finishes at collegiate nationals, across rail jam and slopestyle.
Python, C, C++, JavaScript/TypeScript, SQL
Fusion 360, Raspberry Pi, bus servo control, 3D printing, soldering and power wiring, linkage design
PyTorch, OpenCV, MediaPipe, scikit-learn, Hugging Face Transformers
Eye-tracking and sEMG analysis, electrophysiology, fluorescence image analysis
React, FastAPI, Postgres, Docker
English, Arabic (conversational), Mandarin (elementary)
