Converting natural language questions into SPARQL queries using LLM is much more cost-efficient than using VectorRAG (which might send large amounts of data to the LLM). Metaphactory has this feature ("NL2SPARQL") and it works pretty good for ontologies with good expressivity (proprietary though).
https://help.metaphacts.com/resource/Help:NL2SPARQL
