It'd be a fun project to write a an APL-like that compiles down to NumPy. Would it be pointless? Another interesting project might be to make an APL-like for probabilistic programming.
There's a start for this in Julia[0], where custom string literals mean that you can actually write APL in native Julia:
julia> apl"ι5"
1:5
julia> apl"+/ι5"
15
You can even just get the function that represents the `+/` operation and apply it to any Julia object: julia> apl"+/"(rand(100))
52.17866879741195
0. https://github.com/shashi/APL.jl