The group's 18 clinics span multiple specialties, and each one carried the same hidden shift at the end of the day. Documentation could not happen properly during the visit, because the clinician was, rightly, focused on the patient, so the notes queued up. Doctors then spent close to two hours after every shift reconstructing conversations from memory and shorthand, long after the details had begun to fade.

The costs compounded quietly. Documentation backlogs delayed billing, because a visit cannot be coded and invoiced until its note is closed. Note quality varied between specialties, with each department evolving its own habits and templates. And the human toll was becoming explicit: burnout was climbing across the group, and two physicians had reduced their patient load specifically to keep up with charting, trading clinical capacity for paperwork time.

Leadership faced an uncomfortable trade. Asking clinicians to chart faster risked thinner, less defensible notes. Hiring scribes for every room across 18 clinics would have been costly and hard to staff. And doing nothing meant watching the evenings of its clinicians disappear into documentation, one shift at a time, while the group's most constrained resource, clinician energy, kept eroding.

We designed the system around a simple principle: the consultation itself already contains the note, if something trustworthy is listening. A consent-gated device in the room captures the visit only when the patient has agreed, and speech is transcribed in real time while the clinician stays focused on the person in front of them rather than on a keyboard.

From the transcript, a large language model drafts a structured clinical note mapped to the relevant specialty template, so each department receives drafts in the structure it already works with rather than a generic summary. Nothing publishes on its own: the draft lands in the EHR as a proposal, every AI suggestion is editable, and every change is auditable, keeping the physician as the author of record.

Review and sign-off stayed deliberately central to the workflow. Physicians read the draft, adjust anything the model has phrased imperfectly, and sign the note while the visit is still fresh, typically between patients rather than after the shift. The system's job is to remove the typing and the reconstruction from memory, never the clinical judgement.