[sterlind]

If your project is unlike your normal work it can make sense to use something new.

I was doing some niche amateur cryptography. I started with numpy, but I needed high performance and support for custom fields (e.g. cyclotomic polynomial rings) so I switched to Julia.

The learning curve was worth the expressiveness and conciseness of the new language. It's important that your code matches the abstraction of your problem domain.