I-Corps: Translation potential of a conversational Artificial Intelligence (AI) system for post-discharge patient follow-up

This I-Corps project is based on the development of an artificial intelligence (AI) system that automatically follows up with patients after hospital discharge. Manual nurse outreach cannot scale as care managers make up to 200 calls per day, while most discharges receive no structured follow-up, and the U.S. faces a shortfall of nurses. This technology conducts regular check-in conversations by phone call and text message, reinforcing care instructions, monitoring symptoms, and alerting care teams at the first sign of risk, without requiring an app, patient portal, or technology literacy. Automatic timely follow-up may support discharged patient recovery and free nurses for work that requires clinical judgment. The technology addresses hospital readmission penalties, care-management staffing shortages, and unbilled follow-up care revenue, with applications extending from heart-failure transitions of care to post-emergency-department follow-up, chronic disease management, and remote monitoring programs across hospitals, health systems, and value-based care organizations. This I-Corps project utilizes experiential learning coupled with first-hand investigation of the industry ecosystem to assess the translation potential of a protocol-bound control layer for clinical conversational artificial intelligence (AI). Currently, the control layer is model-agnostic, unlike existing solutions, which are either fully automated but clinically generic or clinically specialized but human-operated. A protocol compiler translates clinical follow-up protocols into machine-readable care journeys, which include executable graphs of scheduled voice and text touchpoints with staged agendas and per-stage permissions. Bounded generation constrains every response to a compiled context of protocol state, curated clinical guidelines, the patient's parsed discharge instructions, and structured chart data, keeping outputs auditable. A dual-channel safety architecture runs deterministic, clinician-defined escalation rules alongside an independent listener agent that evaluates every patient turn, so escalation occurs even if the conversational model fails to act. In addition, structured feedback becomes training data that moves the model closer to how care-managers communicate, engage, detect, and escalate problems with patients. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.