The unprecedented creation, use and share of data around the world has led to new applications and economic opportunities. This data is often large, heterogeneous at a schema and model level, and more or less structured. To bring more order, the World Wide Web consortium recommends sharing data as RDF graphs, which has been mostly adopted in the Open Data initiative, but many other formats are used in practice. Moreover, data is now scattered across places and owners (enterprise silos, open data, Big Tech clouds, etc.) and this adds to the complexity of managing, joining, processing and using various data together. Finally, end users, such as domain experts or decision makers, need tools to generate tangible results they can rely on and share. In this presentation, I will present my PhD and post-doctoral work on this domain. I will first show how to compute structured summaries from any semi-structured dataset. Next, I will show how to homogenize healthcare silos in order to run various federated learning tasks on the underlying data.