Part of the issue is to select the right Wikipedia article. Wikidata offers a way to know for sure that you query the LLM with the right data. Also the wikipedia txtai dataset is for english only.