It is a chain of tools and scripts that analyze the contents of the Wiki and categorize, link and find similarities between one set of content and the next. It does use language models as it needs to read and understand, then convert text into vectors to find similarities. It then relies upon a LLM to stitch everything back together and each of these steps is very compute / storage heavy. The results have been incredible though, thorough and well sourced and cited Wiki articles have been written.
I will say the odds of getting this out today are about 1% however.
I would much rather get something that I know will function with high assurance than ship something riddled with bugs. My issue tracker is near 711 issues tackled. I keep finding these minor things that unfortunately would prevent the site from launching.