Hospitals and independent clinics
Better models and less
admin, on your terms.
We build the record system your clinics run on, and deploy specialist models inside it.
Who it is for
Most clinical AI is trained on a narrow slice of the world: a handful of high-income countries, and mostly patients of European descent within them. A model fitted to that slice does not generalise off it. Accuracy falls on the patients the data left out, which makes it both less fair and, simply, less good.
What your site contributes trains the models your site is given, and we report performance on your own population rather than on a benchmark drawn from somebody else's.
What we offer
An AI-native EHR, and custom models inside it.
Terms
What you get, and what we need.
Where we work
- You get paid
- A contractual share of licensing revenue attributable to your contribution, reported quarterly.
- You keep control
- A veto over onward licensing, and an exit that returns a complete FHIR export with no fee.
- Your current system is fine
- Deployment, contribution and the EHR are separate agreements. We integrate over FHIR or HL7v2.
- We need a clinical owner
- One senior clinician who wants the problem solved and has the standing to stop the pilot.
- We need to publish
- The evaluation gets written up, including a negative result. This is the one condition we will not negotiate.
How a pilot runs
Five steps, agreed before we start.
- I
Scope
One pathway, one metric your department already believes in.
- II
Pre-register
The bar, the comparator and the stopping rules, agreed in writing first.
- III
Shadow mode
The model runs silently beside the existing pathway. No care changes.
- IV
Supervised go-live
Only if the bar is met, with every recommendation overridable.
- V
Publish
Written up either way, then a joint decision about widening it.
Governance
Your data stays under your governance.
De-identification and processing run locally, inside your own environment, under your ethics approvals, privacy policies and national regulations. Identified records never leave the site that created them.
Research consent is sought separately from consent to care and can be withdrawn at any time. A withdrawal propagates to every downstream licensee with a deadline for deletion.
Begin
Bring us the pathway everybody complains about.
The first conversation often ends with us saying a model is not the answer.