Ask HN: ML training in Keras – use Sanic/Flask?

I want to be able to trigger trainings from a web server using a REST interface (POST of course); is Sanic/Flask usable for this? I'd have to run Keras in the background (async job) but I've heard that doesn't work quite well. What is the right pattern to use here? The web seems full of REST calls for inference but not much available for training. If you do this, what's your basic stack?