I’m not the author of the post. Your comment assumes a well known schema. My understanding from the post is that this solution can join and filter on “custom” datasets of arbitrary schema that each of their customers upload.
I've never played with this, but couldn't you create a table based on the dataset that the customers upload, and let your database engine handle filtering those queries? From the looks of it, even if they were doing full table scans for each query, it'd still be faster than all those unions...