Most clinical AI is trained to pass licensing exams, not to hold a consultation. That gap decides what happens when a real patient sits down: whether the model asks the next question or jumps straight to an answer.
How medical models are trained
Medical models usually learn from single-turn question-and-answer data: a symptom goes in, a diagnosis comes out. The result is a model that scores well on board-style questions and falls apart the moment a real patient sits down.
A real doctor does not hear "chest pain" and name a heart attack. They consult. They ask when it started, what makes it worse, whether it radiates, how frightened the patient is. They rule out the common, benign causes before chasing the rare and dangerous ones. The diagnosis is the last step of a conversation, not the first.
Designing the dataset around the consultation
An exam-shaped model fails because its training data has no conversation in it, so we build the data around how consultations actually flow. Every dialogue in Meddies Consultant is generated under three clinical-interview frameworks:
- Calgary-Cambridge sets the arc of the conversation: opening the session, gathering information, building rapport, closing.
- OPQRST forces the AI doctor to work through the symptom one axis at a time: onset, provocation and palliation, quality, region and radiation, severity, timing.
- FIFE shapes the patient side: feelings, ideas, function, and expectations, so the synthetic patient behaves like a person rather than a clean data point.
Each framework constrains what the generator can produce: a dialogue that skips the arc, the systematic symptom workup, or the patient's own concerns does not pass. A model trained on this data learns the shape of a good interview, not just the shortest path to a label.
Why an AI that asks keeps patients safer
A model that answers too fast skips the question that would have changed the conclusion. A model trained on framework-shaped consultations keeps asking instead, because the data never let it reach a diagnosis before working through the symptom. For an ambient scribe (an AI that drafts the visit note) or a triage assistant working beside a clinician, that habit of asking is the difference between a useful colleague and a confident liability.
So we train on the consultation, not the answer key. The model that learns to ask is the model that earns a place beside the doctor.