What's the Most Cumbersome Part of the MLOps Task?

Once you standardize upstream, the downstream stuff should be similar. So once you've got your monitoring processes in place, you understand your retrain strategies and all that,. You can build systems that just automatically do that and scale nicely or whatever. Adam: I think trying to avoid feature stores can because that feels like let's add another tool to cover over the cracks in what we've built from our data landscape. Our data architecture is a bit wonky because it's legacy organically, all this was never concerned. Let's add something else so that we can do MLOps nicely.