I'm building a game using GPT-4 for NPC's dialogue and actions
youtube.com
youtube.com
At the moment I use Unity for much of the game's frontend and core logic, but have a websocket connection to a Node server which handles GPT-4 requests, and connecting to a db. Right now it can be quite expensive at $100/day of R&D, but I'm saving every single request and response, with the goal to finetune a 3.5-turbo or other model in the future, which should significantly decrease costs.
Happy to answer any questions!