Watch a training round where the data never moves.
The whole platform in one loop: the model travels to each institution, learns locally, and only what's safe to share ever comes back.
Three institutions each hold archived imaging. Their data will never move.
The five steps, in words
The model is dispatched. A training job is sent out from the orchestrator to each participating institution — never the other way around.
It trains locally. The model learns inside each institution, against imaging that stays exactly where it lives. No records are copied or pooled.
Only updates return. What comes back is a set of model weight updates — mathematical adjustments, not images and not patient records.
Updates are aggregated safely. The orchestrator combines them under differential privacy and statistical disclosure control, sharply limiting what any output could reveal about an individual.
The result is certified. The finished output is signed as a Verifiable Research Certificate, bound to its lineage and audit trail — checkable by anyone, offline.
Try It Yourself
Now drive the real portal — no account needed.
A ten-minute guided tour of the actual platform on sample data: accept a study invitation as a data custodian, approve a node, open a completed federated study, and verify its research certificate. Nothing you do is saved.