The Neuroscience Informed Learning and Education (NILE) Laboratory at Dartmouth's Geisel School of Medicine operates AI Patient Actor 2.0, an open-access web platform at patient-actor.dartmouth.edu where health-professions trainees interview simulated patients and receive immediate formative feedback. The April 2026 educator manual describes how the system is put together.
The platform is powered by a mixture of large language models that play three roles: they act as the conversational patient, draft clinical cases, and evaluate student encounters. The manual does not name the models. The design principle is that the models do not generate medical content on their own. Every response is constrained by three educator-authored inputs: a Patient Profile, a Configuration, and faculty-defined rubrics and feedback instructions.
A Patient Profile holds the persona: demographics, diagnosis, chief complaint, history of present illness, past medical history, medications and allergies, family and social history, review of systems, personality, communication style, a voice chosen from a voice library, and diagnostic tests with expected values. An optional AI Patient Profile Assistant drafts a full profile from four prompts (diagnosis, goal of the encounter, Big Five personality levels, and free-text communication notes), which the educator then reviews and edits section by section.
A Configuration holds the pedagogy. Its student-visible part gives the context and goal, an ordered list of encounter tasks and optional post-encounter tasks. Its hidden part sets the diagnosis type, role-playing instructions for scenarios that depart from a standard physician-patient visit, encounter feedback instructions that shape tone and detail, optional notes feedback for written work such as a SOAP note, and the rubric, whose categories are each scored independently. Any Patient Profile can be paired with one or more Configurations to form a Case, so the same patient can serve first-year and clerkship learners with different rubrics. Cases are assigned to Courses, and students enroll with a six-character Course Code.
In an encounter the student picks text or emotive voice-to-voice mode and one of 52 supported languages, then interviews the patient. A sidebar carries the task list, a notepad and an Exams and Tests tab for ordering vitals, labs or imaging. On ending the encounter, and submitting a most-likely diagnosis if the Configuration requires one, the student receives rubric scores with narrative coaching; voice sessions add a prosody analysis of how things were said. Cases can be retried without limit.
The platform was designed by Dartmouth students in the DALI Lab and at Geisel and is free to non-profit educational institutions worldwide. The manual reports that deployment at Geisel brought a 76% reduction in late submissions of clinical notes.