Dietary Assessment and Interventions using AI, Wearables and LLM

This talk reviews dietary assessment techniques, including self-reporting, image-based methods, biomarkers, and wearable-based sensing, and discusses their tradeoffs in cost, accuracy, participant burden, privacy, and continuous monitoring. It then presents three AI-driven research directions. The first combines food photographs with post-meal blood glucose patterns through joint embeddings to improve calorie estimation. The second, MealMeter, uses glucose monitor and wristband signals collected during standardized meals to estimate macronutrient intake. The third, MetaPlate, uses free-living physiological and meal data to forecast maximum post-meal blood glucose, applies counterfactual AI to adjust meal macronutrients below 140 mg/dL, and uses LLM-RAG to translate the counterfactual findings into practical meal suggestions. The seminar also covers data collection, model training, feature interpretation, and refinement through registered dietitian validation.

May 5, 2026 12:00 PM — 12:50 PM

T32 Spring'26 Seminar Series