• Developing a privacy-preserving multimodal large language model system for personalized multiple-sclerosis lesion-activity prognosis from clinical narratives and MRI.
• Built a local retrieval-synthesis-critique pipeline with grounded evidence.
• Analyzing systematic failure subgroups using multimodal decision embeddings.

Supervised by Professor Tal Arbel.

• Proposed a fair resource-allocation framework with a probabilistic recommendation prior, optimizing global welfare while preserving user preferences under diversity constraints.
• Developed explainable graph-learning methods for psychosis prediction from conversational text, including speech-graph construction, classification, and graph-frequency interpretation.

Supervised by Professors Laks V. S. Lakshmanan and Xiaokui Xiao.

• Developed MADS, an ensemble multi-agent LLM debate system for political-bias detection, with paragraph-embedding-based debate routing.
• Outperformed supervised-learning baselines and larger LLMs and contributed a dataset for political-article classification.

Supervised by Professors Benjamin C. M. Fung and Elena Obukhova.

• Designed a mobile-device NIR data-collection protocol for non-invasive glucose concentration prediction.
• Applied deep learning to hyperspectral reconstruction and used model explanations to identify informative wavelengths for the sensing protocol.

Supervised by Professor Mohamed Hefeeda.