In many machine learning applications data is sensitive, and we lack concrete methods to provide guarantees that sensitive data is treated with high security and traceability standards. To tackle this problem, Owkin has developed a platform for traceable machine learning on hidden data coming from different sources. The platform, called Owkin Connect, offers strong data privacy, by preventing anyone other than the data owner and chosen algorithms from reading the data. It is based on the underlying open source framework Substra. In a distributed learning approach, data remain stored on the provider’s server and algorithms are trained as they travel between different servers. These computations are orchestrated by a Distributed Ledger Technology, which offers total traceability of operations. Machine learning models, resulting from the training of algorithms on sets of data, belong to both data and algorithm providers according to specific permission regimes. Owkin is currently using this platform in Healthcare to build a network of hospitals and research institutions. It helps creating efficient machine learning models from diverse data sources. In this talk, we will start by detailing the architecture of the platform. We will then present the use cases for which we deploy the platform and lessons learned from these deployments.