Selected publications, research projects, and independent ventures.
Built a full-stack discovery platform mapping 5K+ news articles per run to 100K+ research documents through embedding retrieval, clustering, and batched ingestion.
Unsupervised clustering of 50k+ patient records to infer medical specializations.
Embedded 79,680 biographies with 768-dimensional MPNet representations and analyzed gender patterns through cosine-kNN neighborhoods, 50 randomized baselines, and an interactive searchable 3D explorer.
Built an AI-powered coding platform with a FastAPI execution engine and extensible React 3D runtime that turns accepted algorithm traces into interactive visual explanations.
Contrastive Siamese localization model disentangling disease labels from bounding boxes to improve generalization.
Semi-supervised U-Net segmentation of non-gated CT scans for cardiovascular risk stratification.
Cost-sensitive analysis of error tradeoffs in binary medical ML tasks.
Image-subtraction-based Mask R-CNN for heterogeneous tumor segmentation.
Attention-augmented CNN to compensate for diagnostic variability in CXR interpretation.
Appliance-level inference from smart-meter time series using shape-based clustering.
Exploring how technical ideas scale into real-world impact.