The current tool uses a more complex database primarily designed for the custom search in https://fr.wiktionary.org/wiki/Wiktionnaire:Recherche_avanc%C3%A9e. That tool is way more complex and requires to parse the xml dumps to get data (with a dedicated set of scripts).

Also, note that this random page generator could be replaced by getting a random page in a language category (which all projects have), if it wasn't for the filters I put in place. I excluded the following pages: inflections, names, demonyms, i.e. the numerous pages that were generated by bots and that are so common that a normal search would almost always return one of them. Those are usually not very interesting words and would drown all the other lexemes that we host.

So instead of "copying" the tool, or starting from scratch, we could simply make use of the existing random search in categories, but based on a list of categories (white list/black list) for a given language.