First? What about the crossing by Cecilie Skog and Ryan Waters in 2010? https://www.explorersweb.com/polar/news.php?id=19064

They weren't alone.

Take a look at Vespa. The recommendation use case is featured in one of the tutorials: http://docs.vespa.ai/documentation/tutorials/blog-recommenda...

This looks interesting. Will explore more.

It is more like an alternative to tensorflow serving - it does not train models but is good at evaluating them and use the evaluation for ranking.