Keras Workflow for Experiments

I just reviewed and discarded several packages we used to experiment with neural nets in the late 90s.  Fairly easy to use, but little iterative architecture capability and data wrangling capabilities.    Had been asked about Keras,  an open source neural network library written in Python.  Have never used it commercially.  Also found a piece in KDNuggets on a  Keras 4 step workflow that's useful.  If you are in the early stages of experimentation, and you are already in Python, this makes sense.    If you are beyond that,  something like Tensorflow looks better to test what kind of muscle might be required.