Featured Projects
PyTDC
A multimodal machine learning training, evaluation, and inference platform for biomedical foundation models. I contributed infrastructure to load, benchmark, and fine-tune therapeutic AI models like scVI, scGPT, Geneformer, and ESM, molecule generation metrics, and drug sensitivity prediction experiments.
Astrodash
Website and API for automated supernovae spectrum classification. Hosted under UIUC's SCiMMA, it uses Machine Learning models, like CNNs and Transformers, to classify 50 spectra in under 1 second. Boasts 97% accuracy on test set and capabilities for user-uploaded models, batch classification, redshift estimation, and interactive visualizations. Currently writing a paper for JOSS describing the project.
PropAI
[In progress] Machine Learning Platform that hosts Bayesian-informed machine learning models for predictive analytics across football, basketball, soccer, baseball, and hockey. Leverages the Betfair API to get real-time odds and data, which is then used to find discrepancies with predicted stats and give users the probabilistic edge.
Prediction Market Arbitrage Bot
[In progress] Arbitrage tool leveraging Kalshi's and Polymarket's APIs to identify and trade on cross-exchange price discrepancies.
Pokerbots: Deep Reinforcement Learning for Poker
Developed a Deep Q-Network to learn a poker variant, performed feature engineering and hyperparameter finetuning with cross validation, and integrated into MIT's Pokerbots engine to enable autonomous gameplay. Placed top 10 in final competition.
Minesweeper
Minesweeper implemented in C++ using SFML. Includes GUI, board generation, and game logic. Difficulty ranges from easy to hard, functionality includes flagging, revealing, and game status detection, and dynamic sizing controls the game relative to screen size.