Take it from someone who has to occasionally work on an over complicated type system which uses HM inference, it ain't all that fun.

Could you be more specific? What makes it painful?

(I mean, "overly complicated" is bad news no matter what the specifics of what we're talking about, but how does it cause pain when it's specifically in the types?)

For me personally it's just a lot of really heavy and very "meta" code that is really hard for me to reason about. I also touch it very rarely, so there's always a steep learning curve when I'm jumping back into it.

You could use ReasonML in JS projects. It's even compatible with the rest of your codebase, if you don't want to start by rewriting everything.

Almost all modern statically typed languages use the ideas first coherently described in the Damas-Hindley-Milner approach. The main reason we do not see more full type inference, is that full type inference becomes uncomputable pretty quickly, once you move to more expressive typing systems.

You can go work for Jane street.

Perhaps more achievable: iOS dev with Swift. Not technically HM I don't think, but it functions more or less the same.