Developing a Testable Architecture for a ML Project

We usually separate analyzes from actual code. There is code which we think of as shipping. This is going to be our system. And then there is things like, okay, I don't have a database,. I'm going to do that in pandas. And then I'm just getting the data from this kind of protocol system and then start reading the and then looking at whether it works. Right. Think about the good quality issues of like, are we maintaining this? Are we going into a direction that we come back out? How is it going to look a system? As we progress, it's going to start looking better. But when we are in a state that we say