This is actually the first project I've seen that is actually doing any kind of knowledge graph generation. Most are just precomputing similarity scores as edges between document snippets that act as their nodes. People have basically been calling their vector databases with an index a knowledge graph.

This is actually attempting fact extraction into an ontology so you can reason over this instead of reasoning in the LLM.