Meet Julian — the world's first clinical decision-support AI trained on tens of thousands of real urology cases, curated by a practising NHS consultant surgeon.
General-purpose models are trained on old textbooks, out-of-date guidelines, and scraped internet text. They've rarely seen the pauses, the follow-up questions, the hedged uncertainty of an actual clinic room — the texture that separates a safe tool from a plausible-sounding one.
Most frontier models are trained on poor quality data. Julian has been trained on thousands of real consultations by a practising surgeon.
Tools built without clinical authorship rarely reflect how experienced consultants actually reason and communicate.
Most health-AI products bolt on compliance late. We're building the regulatory pathway in from day one.
Julian started life as a virtual patient, trained on tens of thousands of real urology cases. An experienced clinician (attending) has since consulted with Julian thousands of times — and every one of those transcripts and audio recordings feeds back in, training Julian into the decision-support AI now live in clinic.
Every consultation — real or simulated — feeds the next version of the model. The pipeline is the product.
Real and simulated consultations, recorded with consent.
Speech-to-text tuned for clinical dialogue and terminology.
Transcripts cleaned, labelled, and prepared for training.
A domain model refined on this data, not generic web text.
Deployed as training tools that generate the next round of data.
Julian is built from inside the clinic, not around it. Dr Conor is a practising Consultant Urological Surgeon in the UK, across state and private practice, and co-author of a major textbook in urological surgery.
If you hold an NHS email address, you can request access to a training build of Julian for simulation and CPD use — separate from the clinical decision-support product.
Health-AI credibility is won or lost on compliance. We're mapping our pathway early, not retrofitting it after launch.
Assessing classification under the UK Software as a Medical Device framework as patient-facing features are developed.
Building toward UKCA conformity for any component that meets the medical device threshold.
Data protection impact assessment and UK GDPR compliance built into the pipeline design, not added later.
Early engagement with medical defence organisations ahead of any patient-facing deployment.
We're raising to expand the transcript pipeline and bring the first fine-tuned model to clinic.
hello@opsense.health