Projects
Things I've built and worked on.
Temporal Motion Matching via Transfer Learning
A two-stage deep learning approach using Siamese CNNs and ConvLSTM autoencoders with attention to learn temporal alignment of movement sequences across subjects.
RL Maze Solver: DQN vs Q-Learning
A comparative study of Deep Q-Networks and tabular Q-Learning for maze navigation, with three exploration strategies and a custom Pygame environment.
Taiji Movement Classification
A comparative study of KNN and deep MLP classifiers for recognizing 39 Tai Chi movements from motion capture data, using Fisher's LDA and leave-one-subject-out validation.
Polynomial Regression: MLE vs MAP
A from-scratch implementation of polynomial linear regression comparing Maximum Likelihood Estimation and Maximum A Posteriori inference on noisy sinusoidal data.
Nittany Market
A full-stack e-commerce marketplace built with Flask and SQLite, featuring multi-role authentication, dynamic category hierarchies, shopping cart, checkout, and helpdesk support — developed for Penn State's CMPSC 431W (Database Management Systems).
Mutation Prediction
Benchmarking fine-tuned protein language models (ESM2, ProtBERT) against classical alignment tools for predicting pathogenicity of single amino acid variants.
Voltorb Flip RL
A reinforcement learning agent trained with PPO to play Voltorb Flip, the Minesweeper-meets-Picross minigame from Pokemon HeartGold/SoulSilver.
Wind Wake Optimization
A Streamlit-based GUI for optimizing wind turbine yaw angles using FLORIS, built as a capstone project in partnership with BP.
