You are welcome to joint discussion in kotl.in/slack #mathematics channel. There are several solutions already, but I do not think that following numpy is a good strategy. Kotlin can do much better. @Roman_Belov listed kmath as a numpy follower in talk and Kotlin for DS description, and this was the idea at the beginning, but now I am pretty sure, we can do much better from the point of view of API and at least the same from the point of view of performance, so we are working on something kotlin-ish, not simply a numpy clone. Kmath does not get a lot of love at the moment because I really need feedback from potential users and because I am finishing a project with JetBrains Research at the moment, but any discussion would help (we have a student team who can work on some issues and features).
We are using kotlin in a scientific “production” for two years and are quite happy with it. I personally believe that kotlin is indeed the next language for scientific and non-scientific computations. Especially with the progress made by @Roman_Belov’s team.