Thanks for playing, and the constructive feedback! I created the ~90 puzzles (i.e. the "answers" and the "clues") manually while searching Wiktionary phrases and Google 2-gram data for ideas. Since I already wrote a website that parses that data, the data was already available for me to query. I mainly focused on ambiguous words with the hope that they would have a bunch of phrases (since they'd be used in a lot of contexts).
Thankfully I didn't build the list of partially-correct words manually :D. I wrote code that uses the 2-gram data to build the list of the top 500 strongest connections for a given clue, and builds that into the level data. Hope that helps! TremblingInferno had a similar question so there's a slightly longer response below in case that helps!