all projects

dr.active

Developed a regression model using RDKit and scikit-learn's random forest to predict binding affinity (MAE) from SMILES string and target name with 70% accuracy. Subsequently implemented a Graph Neural Network to increase arruracy by 15%. Used Gradio to deploy interactive demo to web.[demo] [code]

Ripple

Users join a local community, complete skill-based challenges with proof verification, and pass them forward to climb local leaderboards. Built for Minnehack 2025, we ranked 4th[demo] [code]

GopherGuessr

Wrote frontend code for a UMN-based location guessing app using JSX, React, and Tailwind to implement leaderboards, user profiles, and gameplay loop UI.[demo] [code]

BioScrolls

Fine-tuned BioBERT and relation extraction techniques to mine multilingual biomedical literature for neurological gene-disease-drug associations, building a time-indexed knowledge graph with key metrics such as entity frequency, relationship confidence, and trend analysis over time.[demo] [code]

dr.optimus

Developed a drug optimization pipeline using Deep Q-Networks (DQN), where the agent modifies molecular structures represented as graphs using a Graph Neural Network (GNN)-based environment; evaluated optimization success using QED (Quantitative Estimate of Drug-likeness), binding affinity scores from docking simulations (e.g., AutoDock), and toxicity predictions generated by pre-trained classification models.[demo] [code]

folduzz

Developed a deep convolutional generative adversarial neural network to predict protein folding.[demo] [code]