Reliable learning systems for changing, high-stakes worlds.

Maternal safety & perioperative intelligence

At the Kovacheva Lab, led by Vesela Kovacheva, MD, PhD, I contribute to clinically grounded predictive modeling for maternal safety at Brigham and Women's Hospital and Harvard Medical School. The goal is to surface risk earlier and support safer care through models that combine the full clinical picture rather than relying on a single data stream.

• Multimodal signals: electronic health records, physiologic waveforms, genetic and imaging information, and clinical text.
• Clinical questions: hypertensive crises, hemodynamic instability, hemorrhage, and intensive-care admission.
• Design principle: clinically meaningful predictions that are robust, interpretable, and evaluated for real-world use.

Making narratives useful for forecasting.

My doctoral work developed multimodal approaches for modeling longitudinal experiential data. I study how language models can encode reflections, open-ended responses, and other qualitative signals alongside numerical trajectories to predict meaningful change.

• Representations: narrative-based language-model features for temporal and experiential data.
• Outcomes: behavioral and emotional engagement, confidence, science identity, and academic performance.
• Reliability: cross-distribution generalization, missing-data handling, and transparent evaluation across diverse populations.

Applied intelligence across domains.

My earlier research spans agricultural sensing, computer vision, human activity recognition, health monitoring, and educational AI. Across each setting, I am interested in moving from promising model accuracy toward systems people can understand and trust.