@dehora And re: 3), I think the first phase was devops ideas being applied to the parts in the ML workflow where they fit more or less. It is not bad but also not great.