Personal health data is widely available and frankly overwhelming, especially from continuously sensing technology like wearable devices. However, users are not afforded the ability to ask free-form questions about their health data to discover hidden insights, trends, and patterns. The majority of current solutions available to support users with learning from their health data present fixed summaries and trend plots that are not sufficient to answer targeted questions that a user may have about their own health trends. For example, a user might want to know "how did my sleep duration change in the summer months compare to the winter months?" or "when was my glucose least stable in the past month?" With current solutions, a user needs to download their own data and then manually calculate the metrics needed in an attempt to find answers to their questions. While this might be feasible for a select few, the current solution is not feasible or sustainable for many.

To address these gaps, the Health Data Question-Answering project seeks to investigate opportunities around using large language models as interpreters for natural language queries on personal health data. ‍

This project is co-led by Dr. Nikhil Singh & Dr. Temiloluwa Prioleau in collaboration with Dr. Ali Emami.