Awesome - glad that the folks at the big G are tackling this problem. IMO the field is in sore need of tools that help open the black box - I think it's going to be a big challenge to design such tools in a way that they are both flexible and powerful[1], but composeable building blocks certainly feel like the right start.

[1] This is a big challenge for DL tools in general - if what you're trying to do can't be expressed out of standard tensorflow ops, you're going to have a bad time.