We started by researching privacy-preserving AI: how to train and use models without collecting people’s data in one place. That work became Hura, a federated learning prototype, and in 2025 the project was selected among more than 770 projects for the AI Action Summit in Paris.
Talking to banks, clinics and security teams, we heard the same problem again and again: they want to use AI, but they cannot send their data to it. So we turned our research into a product: a private AI gateway that keeps sensitive data inside and produces the evidence European rules ask for.