The factory for cloud-edgeization: turning local apps into cloud-edge AI. Federated training, sovereign inference, AI-Act-grade governance, pre-integrated across European borders. Compose a system. Ship it. No data leaves.
Born inside the €3B IPCEI-CIS as its WS3 horizontal reference, Ambiti8n exists to prove one thing: that the work stream's partners — eight companies across four countries — can interoperate across the cloud-edge continuum, on the 8ra sovereign cloud. We proved it live with our first vertical: sleep apnea detection, federated across hospitals in Italy and Poland. The next verticals won't start from zero.
Most European AI still lives where it was born: a single hospital, a single plant, a single dataset. Ambiti8n is the assembly line that lifts those local apps into the cloud-edge continuum: federated training across sites, sovereign inference at the edge and in the cloud, with AI-Act-grade governance pre-integrated end to end.
Six interoperable steps, each owned by specialist European partners. Hover to pause, or use the control.
The foundation: two hospital zones across Italy and Poland plus a federated AI Factory, running on Kubernetes clusters and virtual machines from three providers. Each zone is self-contained. No patient data crosses a border.
Docker image → Helm chart → TIM registry → deploy anywhere. One package runs consistently across heterogeneous clusters. Kafka passive monitoring adds zero overhead, so a clinician is never the first to discover a failure.
A developer picks the app from the Service Catalog and clicks deploy. TIM's Service Orchestrator places containers on the right clusters, wires cross-provider networking, and activates monitoring. From user action to live endpoint in seconds, not hours.
At each hospital: openEHR database → data pipeline → local FedX storage, harmonised through the OMOP common data model. The data never leaves the building. The service travels to the data, not the other way around. Reply's Kafka-based monitoring watches the pipeline end to end, so failures surface before they reach a clinician.
The FedX client trains locally at each hospital and sends only gradients. The FedX server at Gdańsk TU coordinates rounds and aggregates them. Gradients are mathematical derivatives, not patient IDs. Convergence is tracked across every federated node.
Path 1 - edge: FedX Inference runs the CNN on hospital hardware, low latency, audit-logged locally.
Path 2 - sovereign cloud: the LLM runs on Sherlock in the EU. Engineering's AI Model Catalog connects training output to the Sherlock deployment.
The foundation: two hospital zones across Italy and Poland plus a federated AI Factory, running on Kubernetes clusters and virtual machines from three providers. Each zone is self-contained. No patient data crosses a border.
Docker image → Helm chart → TIM registry → deploy anywhere. One package runs consistently across heterogeneous clusters. Kafka passive monitoring adds zero overhead, so a clinician is never the first to discover a failure.
A developer picks the app from the Service Catalog and clicks deploy. TIM's Service Orchestrator places containers on the right clusters, wires cross-provider networking, and activates monitoring. From user action to live endpoint in seconds, not hours.
At each hospital: openEHR database → data pipeline → local FedX storage, harmonised through the OMOP common data model. The data never leaves the building. The service travels to the data, not the other way around. Reply's Kafka-based monitoring watches the pipeline end to end, so failures surface before they reach a clinician.
The FedX client trains locally at each hospital and sends only gradients. The FedX server at Gdańsk TU coordinates rounds and aggregates them. Gradients are mathematical derivatives, not patient IDs. Convergence is tracked across every federated node.
Path 1 - edge: FedX Inference runs the CNN on hospital hardware, low latency, audit-logged locally.
Path 2 - sovereign cloud: the LLM runs on Sherlock in the EU. Engineering's AI Model Catalog connects training output to the Sherlock deployment.
The people building Ambiti8n, on what they're seeing emerge.
The same assembly line is ready for the next vertical. Swap the data source, swap the model. The federated training, governance, and sovereign serving stack stays the same.
Sleep apnea detection across hospitals in Italy & Poland. Federated, sovereign, live.
● Live · System #01Predictive maintenance across multi-vendor factory floors, without exposing process data.
Vertical readyAnti-fraud and AML models across European banks, without pooling customer data.
Vertical readyLoad forecasting across national grids, preserving operator sovereignty and data residency.
Vertical readyAmbiti8n is open for new consortium partners, pilot hospitals, and sector leads across manufacturing, finance, and energy.