Now I spend most of my spare time working on a little recommender
system engine implemented with Tensorflow, and particularly
Tensorflow Serving. There are quite many other realisations of
different recommending algorithms in Tensorflow or PyTorch, but
usually, they are not production oriented and don't include any
serving submodule which is a common point for research projects but
which is definitely intolerable in any real business application. I
can already say that Tensorflow Serving is really fast, although I'm
still writing and grooming the code.
