It's a bit different: take a dataset classified according to Schema’s ontology. These individual objects can be indexed based on an embedding. But if you do that, you don't need the formality anymore. Weaviate is schemaless by default but you can enable it if you like, and it exactly fits
Schema.org, there is a repo floating around somewhere with the code to do this too. I was a big proponent of ontologies, semweb, etc but made a 180 after the embeddings entered the scene and never went back