GlucoFM

GlucoFM is a predictive model that looks only at the glucose signals measured every few minutes by a wearable continuous glucose monitor, and from that alone flags early signs of conditions like diabetes risk or insulin resistance.

It helps to think of the glucose signal as a river. Throughout the day there's a big, slow-moving current, and on top of it are small ripples caused by things like eating a meal or moving around. Older approaches lumped this river together as one signal, unable to tell the big current apart from the small ripples. GlucoFM instead splits the signal into two streams from the start, examining the big current and the small ripples separately before combining them to make its judgment.

The model learned its patterns without any answer key like hospital diagnostic records — it trained on huge amounts of raw numbers straight from glucose sensors. It was also deliberately exposed to real-world messiness during training, like sensor dropouts and erratic spikes, so it holds up well even when actual wearable data is noisy.

In articles it appears as the name of a newly released model, as in "Google Research... unveiled the foundation model GlucoFM on August 26." A common misunderstanding is that GlucoFM is not a medical device that instantly diagnoses diabetes. It's a research model that estimates risk purely from glucose signal patterns, without hospital labels — actual clinical diagnosis still requires separate testing and a doctor's judgment.