implementation is not done yet, it's more about the proper mental model for generalizing numpy
chatted with stephen about Repa and hpc code engineering the evening before the talk, its all haskell ideas ported to python land under the covers :)
(as the slide deck admits!)
Well, that's not entirely true. It's based on many of the same ideas which underlie some of the good Haskell ideas, and have a natural expression in Haskell because it's a type algebra on index spaces.
An incomplete list of the inspirations: Paralation Lisp, Merrimac, APL / J / K, Manticore, FORTH, Chapel & ZPL (and VCODE), SISAL, OLAP & MDX, LINQ, various PGAS approaches, etc., etc.