But I recently picked up a "Machine Learning" project for a client. I don't entirely understand everything in the fullest detail, but I understand the basics. Yet despite my lack of knowledge I got something to work 'enough' in a day. Tensorflow makes getting a working DNN easy, and then wrappers like TFLearn make it nearly trivial.
The hardest part of the project was sourcing a decent amount of training data.
And I'm honestly somewhat ashamed of it. My friends think it's cool, but really I don't feel like I did anything. I just glued a library together.
But maybe that's the interesting bit, that frameworks have been developed to a point that even a rube like me can get something working in a day.