The Python code using C libraries you get on machine learning is a nice model. There is no reason why you can't recreate it with Haskell + Rust.

But it has one drawback that it's relatively slow to pack stuff to send between languages. On Python everything is so slow that this doesn't make much of a difference, but on the Haskell + Rust scenario you will want much larger library blocks.

You will lose the benefits of both type systems if you try to combine them. Rust doesn't really track effects, while Haskell doesn't care about memory.