Controlled benchmark testing no-training multimodal retrieval baselines before learned alignment.

AI/ML research and software portfolio focused on multimodal learning, retrieval systems, representation learning, medical AI, and model evaluation.

Controlled benchmark studying whether DANN domain adaptation improves retrieval or damages embedding neighborhood structure.

Research portfolio connecting my work on multimodal learning, retrieval systems, contrastive learning, embedding geometry, and AI evaluation.

Collaborative-filtering recommender built from scratch — Truncated SVD + SGD-trained bias terms in NumPy, with embedding probing for implicit demographic signal

Controlled benchmark showing how spectral geometry diagnostics reveal embedding failures hidden by retrieval metrics.