What is true today
Clinical AI is tested mostly on exam-style questions and on records from a small number of Western health systems. Most of the world's care looks different.
In a growing number of countries, health data cannot legally leave.
And the leading AI labs have started working in these regions: a $200M partnership for health and education that includes benchmarks for sub-Saharan Africa and India,1 a $50M programme to bring AI to 1,000 African clinics,2 and a study of a clinical assistant across 39,849 patient visits in Kenya.3
What we believe
Models will need to prove themselves on real clinical work in the places they are deployed.
The data for that will not be exported, so the testing will have to go to the data.
The health systems that hold it should be paid, and should stay in control. Someone neutral has to build and run the environments. We intend to be that company.
What we are building
Environments inside health systems in emerging markets, with a model that shares revenue with the data holder.
Our first environment is for tuning models on specialist care. It is the first of many workflows, and it is how we are learning to build inside a health system properly.
An invitation
If you build models, hold clinical data, or invest in this field, we would like to talk.