Sovereign inference starts with knowing what you're running.
Ambiti8n is a federated AI platform built under IPCEI-CIS, bringing together technologies from eight companies across four European countries. As part of the initiative, Engineering focused on ensuring that AI models deployed in healthcare environments remain trustworthy, verifiable and compliant throughout their lifecycle.
The platform solves a problem that stops most healthcare AI projects before they start: how do you train and deploy AI models on sensitive clinical data that cannot leave the hospital? The answer is federated learning. But federated learning only solves half the problem. Once a model is trained, a second question immediately follows: how do you know that what runs in production is exactly what was validated?
This is the question Engineering took on.
Engineering's role
Our contribution to sovereign inference focused on connecting validated AI models to deployment in a trustworthy and traceable way. This matters because Ambiti8n aims to run clinical AI inference on sovereign European infrastructure, with patient data staying inside the hospital. For such a high-risk domain, model management cannot be just a folder or a version number: it must provide governance, traceability, and confidence in what is running in production.
Engineering's delivery helps make sovereign inference not only technically possible, but also explainable, auditable, and trustworthy.
The hardest part
The hard part was not building the AI inference itself. The real challenge was making sure we always know which model is being used, where it comes from, and whether it has been properly validated.
In healthcare, AI systems are considered high-risk under the European AI Act. That classification carries real consequences: trust, transparency, and traceability are not optional extras, they are legal requirements. If you cannot explain why a model is running and whether it has been approved, you are not operating a clinical AI system.
You are operating a black box. Our job was to make sure that never happens.
The AI Model Catalogue is Engineering's answer to that challenge. Every model produced by the federated training pipeline receives a clear identity before it reaches the inference platform: where it comes from, which version it is, whether it has been validated, and whether it is suitable to run in a clinical context. Nothing reaches production without going through it first.
What sovereign inference really means
There is a tendency to define sovereign inference purely in terms of geography which cloud, which country, which jurisdiction. And that matters. But it is not enough.
Sovereignty without traceability is fragile. You can run a model on European soil and still have no clear answer to the questions that matter most in a clinical context: Is this the right version? Was it validated for this use case? Can this decision be audited?
Engineering's contribution addresses exactly that gap.
In the end, sovereign inference is not just about where the model runs. It's about knowing what it does, why it does it, and that it can be held accountable. In Ambiti8n, Engineering made sure that trust is not assumed - it's built in.