AI Virtual Nurse for Patient Engagement

The problem

The window after discharge is where recovery quietly goes wrong. Patients miss medication, symptoms escalate unnoticed, and the first signal anyone gets is a readmission. Following up with every discharged patient by phone is the obvious answer and the one hospitals cannot staff.

How it works

The assistant checks in on a schedule tied to the care plan, asking about symptoms and adherence in plain language. Responses are assessed against the patient's own clinical context — pulled over FHIR from the hospital record rather than treated as isolated answers — so a reading concerning for one patient is not flagged for another where it is expected.

Azure Health Bot provides the conversational layer within the hospital's compliance boundary. PostgreSQL holds the longitudinal check-in data, which is what turns individual responses into a recovery trajectory clinicians can look at.

What shaped it

The system's job is escalation, not diagnosis. It is designed to notice deterioration and route it to a person, and deliberately does not offer clinical advice — the failure mode of an over-confident automated nurse is severe enough that the boundary was drawn conservatively.

Missing a genuine deterioration is far costlier than a false alarm, so thresholds were set to over-refer rather than under-refer.

Outcome

Readmission rates fell and adherence to care plans improved, with clinical staff alerted to at-risk patients while there was still time to act.