MDwAIstScheduler brings on-device voice documentation to clinical practice

Clinical documentation forces physicians to split their attention between patients and keyboards, with much of the work spilling into uncompensated after-hours time. Researchers Diego Mardian and Frank Liu address this with MDwAIstScheduler, a low-cost, belt-worn pipeline that lets physicians speak naturally during encounters while structured data lands in the EHR. Building on an earlier cloud-based prototype, the new system runs both speech recognition and intent extraction entirely on-device, using a medical-domain automatic speech recognition (ASR) model and a 1.7B-parameter language model fine-tuned for clinical action extraction.

The pipeline captures medications, allergies, labs, orders, referrals, follow-up scheduling, vitals, and problems as review-ready draft notes directly in the Elation EHR, so the clinician merely confirms rather than types. Because no patient audio or text ever leaves the device, it offers strong privacy protections and addresses HIPAA concerns inherent to cloud-based solutions. The result is a tool that removes keyboard work from the visit without removing the clinician from the record. For overworked physicians, this could mean shorter after-hours charting and more face-to-face time with patients, while maintaining complete and accurate documentation.