One agentic AI for a diabetes population-health operation. On today’s members it senses risk, understands who is on a validated deterioration path, plans each member’s next-best outreach, acts — autonomously where safe — and observes outcomes that feed back. An ML layer (Learn) trained it on 36 months of real outcomes. Synthetic data (no PHI); every number reproducible.
SenseUnderstandPlanActObservetrained by ML · Learn
Source data — the ML training cohort (3,500)
The ML layer (Learn) learns from a bundled, frozen synthetic cohort — no PHI, identical on every run. The live agent then works today’s 5,000 current members. Take a look at the raw training records before we load.
First load trains the ML and starts the agent in ~60–90s (a progress screen shows), then the live view opens on today’s members.