GlucoAgent uses LLM-based code generation for personalized glucose forecasting
November 4, 2026
GlucoAgent employs an LLM-based agentic system to write and execute code against multimodal data like CGM and meal logs at inference time. On the CGMacros dataset, it outperforms time series foundation models across all forecasting horizons by dynamically retrieving past meals and fitting physiological models.
HOW THIS AFFECTS YOU
●
researcherThis demonstrates the utility of LLM-based code generation for complex time-series forecasting tasks.
●
healthYou can use agentic reasoning to handle highly individualized physiological data for more accurate patient monitoring.