Building the patient-understanding layer for AI-native dentistry.

Everyone wants a dental model. Almost no one wants the unglamorous part: datasets where every label was placed by a clinician who will stand behind it. Dentistry is dense with edge cases — restorations that mimic decay, anatomy that mimics pathology, imaging artifacts that fool confident models. Labels scraped from messy records inherit every one of those confusions.

Molar Town is designed as a capture environment: identical suites, standardized imaging, consistent positioning, consent-first records with patients holding their own data. Controlled variables in, trustworthy labels out. That’s the dataset a frontier lab can actually build on — and it’s why the practice and the research are one project, not two.

A tool that can’t write to the record can’t corrupt the record. Read-only isn’t a constraint I apologize for — it’s the safety posture that makes a pilot easy to say yes to. Every DentalDNA output is clinician-reviewed before a patient sees it, and every summary is auditable against the source record it came from.

It’s also a philosophy about how systems should earn trust: the same way people do. Be useful for a long time without asking for more access. The fastest path to a broader integration is a narrow one that never causes a problem.

If you want robots working near humans, you need environments that are bounded, repeatable, and supervised — where the stakes are real but the variables are controlled. A dental operatory is exactly that: a small room, a fixed instrument set, procedures with defined steps, and a trained clinician in the loop at all times.

Six identical suites means six copies of the same experiment. I’m writing the support workflows and safety cases now, before any robot exists in the room — because the safety case should always ship first.