Experimental systems for multimodal learning, biomedical language models, explainability, and evidence-driven evaluation.
A four-class model that separates Alzheimer's disease, frontotemporal dementia, Parkinson's disease, and healthy controls using structural MRI and resting-state EEG together, and keeps working when a patient has only one of the two.
Status · Completed
Does fine-tuning a general LLM on biomedical facts actually reorganise how it represents biology? QLoRA-tuned Llama-3, Mistral, and Phi-3 on 68,444 gene–disease associations, then measured both accuracy and the geometry of their embeddings.
Status · Completed research project
One network that both outlines a lesion in a breast ultrasound image and classifies it as benign, malignant, or normal, trained on the public BUSI dataset.
Status · Completed coursework project
A full machine-learning pipeline over the PHL Exoplanet Catalog (5,600+ planets, 100+ features) to predict habitability class, with explanations for individual predictions.
Status · Completed coursework project
A survey study of how gamified features (rewards, challenges, progress) affect mobile banking use among Bangladeshi university students, tested with four hypotheses.
Status · Completed research project
An end-to-end pipeline over 114,002 Spotify tracks that defines 'viral' by clustering and compares five classifiers, and a useful lesson in target leakage.
Status · Completed coursework project
A stacking ensemble for forecasting U.S. recessions, interpreted with SHAP to check that its signals follow economic logic.
Status · Completed coursework project