PROJECTS
Personalized Causal Graph Reasoning for LLMs: An Implementation for Dietary Recommendations
This project develops a framework that empowers large language models to reason over personalized causal graphs built from longitudinal health data. Each graph encodes how individual factors influence outcomes, allowing the LLM to trace causal pathways, rank their impacts, and simulate potential results. Applied to nutrient-oriented dietary recommendations, the system tailors food suggestions for improved glucose control, addressing variability in personal metabolic responses. Counterfactual analysis and LLM-as-a-judge evaluations show stronger personalization and effectiveness, paving the way for safer and more adaptive health guidance.