That's really nice! One thought - it might be helpful to have some way to seed the system with what you at least think is "normal" consumption?

The one other thing that springs to mind is having some way to mark vacations or other disruptions to schedule? Also potentially a way to mark that you refilled some other way? Often when I'm on a trip, I'll bring home some coffee from a local roaster, so having a way to feed that into the system would be nice.

Interesting concept for seeding the system. We used to have something like that and dropped it because it wasn't too accurate.

We do have a setting for turning off automatic re-ordering to handle vacations or unexpected bounties of coffee.

I like this idea, and good luck! Just wanted to give some input on this as I currently work on an ML project at Amazon. The more context possible from the user to define the space at the very beginning will increase accuracy exponentially (at the risk of a too much effort required for user onboarding). Finding that balance is important. Additionally, the user feedback is very important as well. incorporating a way to say yes a shipment came at the wrong time, or no it didn't will also improve model development. However both experiences should be lightweight and seamless - good luck!

Thanks for the advice! I agree there's a tradeoff between asking the user too much and expecting the system to learn on it's own.