Sometimes, I find myself feeling deep appreciation for well-designed libraries. This happens every time I have to write my own matrix manipulation code. The actual mathematics is not too difficult, and neither is the implementation (optimization aside), but making a good interface is a pain. I will always appreciate numpy and how seamless it is to use the ndarray. The entire library is a great example of an extremely performant backend and a stupid simple interface. Making such a powerful tool so easy to use will always impress and inspire me.
I like numpy :)