Ask HN: How do you perform online learning for Production ML systems?

My workflow for batch/offline training takes a day. I have a use case where I want to aggregate feedback from last hour and perform online learning.

How would you do it in Production ML system serving live traffic?