U-M researchers receive Laude Moonshots funding for AI-driven glucose management

A research team at the University of Michigan has received an Honorable Mention in the inaugural Moonshots competition, a flagship initiative of the Laude Institute. Their project, titled “An AI-Enabled Conversational Advisor for Individualized Glucose Management,” aims to bridge the gap between glucose monitoring and patient decision-making by leveraging large language models (LLMs) to interpret complex physiological data.

The multidisciplinary effort is led by Jenna Wiens, associate professor of Computer Science and Engineering (CSE) and co-director of the U-M AI & Digital Health Innovation initiative, along with co-PIs Nikola Banovic, associate professor of CSE, Irina Gaynanova, associate professor of biostatistics, and Joyce Lee, professor of pediatric endocrinology at Michigan Medicine.

The Moonshots program was designed to fund solutions to “species-level” problems, challenging researchers to apply AI to address humanity’s greatest challenges. Judged by a committee of top leaders in the field, including Turing Award winner John Hennessy and Google Chief Scientist Jeff Dean, the U-M proposal was selected from a field of 125 proposals representing 47 leading institutions and recognized for its potential to transform the daily lives of those with diabetes.

The project addresses a critical challenge in chronic disease management. While continuous glucose monitors (CGMs) provide a wealth of data, patients often struggle to interpret this information between clinic visits. To address this gap, Wiens and her collaborators are developing a conversational diabetes coach that uses LLMs to provide real-time, individualized support to patients.

Unlike standard AI models, the team’s system is designed to reason over dense physiological waveform data and proactively ask follow-up questions to gather necessary context, such as recent exercise or unlogged meals. By combining sensor data with patient dialogue, the advisor can offer actionable recommendations grounded in clinical safety guidelines.

Beyond the conversational interface, the team is building a new cause-and-effect benchmark for AI reasoning. Using an FDA-approved simulator, they are creating a foundation for evaluating how effectively AI models can identify the underlying drivers of glucose fluctuations.

By moving beyond simple data tracking toward proactive, conversational intelligence, this work seeks to lower the cognitive burden on patients and establish a new standard for safe, AI-assisted healthcare. Through these innovations, the team aims to provide a blueprint for applying adaptive AI to a wide range of chronic conditions.