She envisions a future where AI can provide personalized insulin recommendations tailored to individual lifestyle patterns, significantly improving T1D management.

Currently, T1D patients face challenges in manually estimating carbohydrates, often leading to inaccuracies and poor blood glucose management due to the complexity of food nutrition information.

In simulations with 20 virtual T1D patients, the GPT-4o mini LLM demonstrated superior performance over human dietitians in carbohydrate estimation accuracy, achieving 69.82% of the time in the safe glucose range.

Qin is actively fine-tuning LLMs to function as 'LLM nutritionists,' aiming for more precise nutrition estimates for T1D patients.

NutriBench is being utilized to benchmark large language models (LLMs) for carbohydrate estimation tasks, which is crucial for T1D patients.

Yao Qin, a PhD and type 1 diabetes (T1D) patient, is leveraging artificial intelligence (AI) technologies to enhance blood glucose control.

To support this vision, Qin's team developed NutriBench, a comprehensive natural language meal description database containing 11,857 annotated meal descriptions from 11 countries.

Qin's project also aims to develop algorithms for static and dynamic activity-specific insulin presets, which will help mitigate hypoglycemia risks during exercise, integrating with the FDA-approved Tidepool Loop app.

With the use of LLMs, patients can input meal descriptions and receive immediate carbohydrate estimates, enhancing both accuracy and efficiency.

Research indicates that different physical activities affect glucose levels variably, underscoring the need for tailored insulin adjustments during exercise.

The Helmsley Charitable Trust is funding Qin's project focused on automated insulin delivery (AID) systems to assist T1D patients in maintaining stable glucose levels during physical activity.

At the recent AI in Healthcare Virtual Summit hosted by the Endocrine Society, Qin highlighted her efforts to simplify carbohydrate counting and improve exercise management for T1D patients.