In the afternoon at AI-Bridges, Philippe Saadé from @wikimediaDE discusses the #Wikidata vector database and use cases for RAG, named entity recognition/ disambiguation and classification, but not only. The vectors are calculated per language (English, German, French & Arabic), but the search query entered in the LLM interface can be multilingual.

https://wd-vectordb.wmcloud.org/

A complementary development is the Wikidata MCP with a set of available tools. The connection to LLMs like Mistral (where connection to the internet can be interrupted, so that only source of truth becomes Wikidata), seems promising but not error-free yet esp. with SPARL query formulation & need to remove a lot of assumptions.

https://www.wikidata.org/wiki/Wikidata:MCP

And brand new (since 2 days): Wikibase MCP -- https://wb-mcp.wmcloud.org/