Projects
Othello Reinforcement Learning Model
Trained a deep learning agent using TensorFlow, CUDA, and cuDNN for self-play optimization on GPUs.
Receding Coastlines
As part of Shellhacks 2025, three teammates and I built a coastline recession prediction website. My contributions included developing, tuning, and training the transformer model used for prediction, as well as creating the backend component that converts model output into a suitable format. Website
Underwater Submarine Pose Detection
Designed a computer vision pipeline to estimate the pose of a United States Seawolf-class submarine in shallow, overhead-lit underwater-to-underwater conditions. To improve data control and coverage, I built an underwater simulation for training data generation instead of relying only on real imagery.
I first used MASt3R for point correspondences with PnP + RANSAC for pose estimation, then diagnosed feature extraction failures caused by underwater distortion, discoloration, caustics, and bubbles. I integrated Segment Anything 2 (Meta) to segment out background artifacts before matching, which restored robust correspondences and enabled accurate final pose estimation.
ASL Character Classification App
Built a CNN for ASL character recognition (~95% accuracy) and integrated it into a Flutter app using TensorFlow Lite.
Inverse Kinematics Robot Arm
Simulated a robot arm using Jacobian-based IK with obstacle avoidance and joint constraints in Python.
Lecture: CNNs for MNIST
Hosted a lecture + lab on CNNs and the MNIST dataset. View Slides
