What I hope is to be able to create a prototype and see if anyone is interested.
To state the obvious.
https://itch.io/library/recommendations
Here are some things you might like based on your rating, purchase, and download history.
Itch has such a recommendation system and Itch has more data available than you.
Steam has a lot more data. It knows which games you actually play. For a proof of concept Itch is a terrible place to get test data and correlate the fingerprint of a game to how people like it or not. You just cannot use tags in a reliable way here. They do not accurately describe the games. Doing maths on those is acting on false premises. You will not find the "hidden" gems, you will find the "obvious" gems - which people already found themselves or get in their Itch recommendations.
If you have test users, try using their public Steam profile, if they have such. The games on Steam do have mostly accurate tags, because they are user chosen and even ranked by relevance. And they are from a fixed pool.
And then compare your predictions to the Steam recommendations those test users get.
I’m not a data science guy
Steam has those, and I hope Itch asked someone that knows this stuff. The recommendation systems in place on those platforms are made by expert professionals.
There are also independent recommendation systems on the net. This is a thing where research went into. Basically every store has a vested interest in analyzing such data and using it for advertisement recommendations. https://duckduckgo.com/?q=game+recommendation+engine&ia=web