Former Wikia engineer here too! I also thought there was a lot of potential there. We even invested in some RDF and structured data and NLP projects (second screen, sentiment analysis on comments for detecting flame wars, etc), but for various reasons they just didn't work out beyond hackathons and demos. I think there were a lot of well meaning engineers who wanted to make stuff like that work. Part of the problem is mediawiki itself. A page is literally just text using an awful hacked together xml parser and some regexes to emit HTML. It might look like a database sometimes when it is rendered (and there is Wikidata) but there is no actual structure to it, just a pile of templates made of other templates that people have to tediously wrangle by hand. That it eventually turns into some HTML that you can view is almost an accident.