I added a basic roguelike loop to codenames and made it a 2 player co-op game.

The "AI" (if you can call it that... I don't think it qualifies for the "I" part) is based on word embeddings. All logic runs on the frontend, with backend acting as a proxy between two players.

I had a theory that sense2vec (contextual embeddings) may perform better than standard embeddings, but like prior basic codenames AI approaches, it does very well played against itself, but provides incomprehensible clues for humans.