Seeking suggestions for improving a Card game implementation

[mudont]

To improve Functional Programming skills, I decided to implement (Git Repo) the card game of Indian Rummy.

I would welcome any comments on making the code more idiomatic. I struggled a lot to come up with non-imperative code, often using the first functional pattern that worked.

It is a work-in-progress, but one non trivial function works - Given a Hand of 13 cards, produces all plausible melds in order of points (less is better). There is minimally tested code in src/game.ts to model the game.
The goal with this repo is to flesh out simulate_game.ts so that a few types of bots with different strategies can play against one another, and observe their success rates. I am not sure I can go as far as using reinforcement learning to improve strategies.
I am separately working on a React front and and a node backend so my friends/family/anyone can play this game online. It will all be open source of course.
Probably the bigger goal is to build it into a good real world fp-ts application. I definitely would need a lot of advice to make it a good model of functional and strongly typed code.