GlucoPulse: An Open-Source AI Metabolic Guardian I Built for My Friend Alex

This article is about an open-source AI tool called GlucoPulse, created by the reporter for their friend Alex. Alex, a 24-year-old software engineer, was diagnosed with pre-diabetes after receiving an HbA1c reading of 5.9% from routine bloodwork. Feeling overwhelmed and anxious about their blood sugar spikes, Alex bought a Continuous Glucose Monitor (CGM) but found mainstream apps overwhelming and data-harvesting. Alex wanted to know how specific foods would affect their blood sugar before consuming them.

GlucoPulse is an open-source, privacy-focused digital twin designed specifically for Alex. It takes the meals Alex enjoys, combined with physiological biomarkers, and runs them through Prior Labs' TabPFN model to forecast postprandial peak glucose, 90% Bayesian confidence intervals, spike risk severity, and 180-minute glycemic curves. The tool also includes counterfactual what-if simulations showing the effects of taking a walk, adding fiber, or sequencing foods differently.

The reporter handed GlucoPulse to Alex with a spicy Thai Pad Kee Mao noodle meal and Alex's history of sleep deprivation. The tool predicted a sharp glucose spike and demonstrated how eating the meal with a walk and a salad could reduce the peak glucose level. Alex was pleased to learn that they didn't have to give up their favorite food entirely, just change the order of consumption and add a walk.

The reporter believes GlucoPulse gives Alex peace of mind by running on their own machine, without subscriptions or medical tracking.