Thanks for the tips!
133 karma · joined May 25, 2015
Master's AI student at Charles University in Prague. Former intern at Dropbox, Google, Prizeo.
I don't have a big minizinc input on hand. In our database it takes about 2 minutes to plan for a weeks meals. The model is probably wildly inefficient and optimizing it is one thing on the long term bug list.
It schedules meals for my shared household. I can give it a request like: "next week I want to cook 4 vegetarian meals with 3 portions, and 1 vegan meal for 8 people for big shared dinner, and I want <=2 of those to be noodles.
It thinks for a bit with a planner (implemented in Minizinc - https://www.minizinc.org/ - with OR-Tools for a solver), and out pops a plan, like: "OK, you need to check you still have 500 g of potatoes, ..., then buy 1 kg of carrots, 1 liter of coconut milk, and here's links for the online store; I also checked those ingredients in stock; I also need cayenne pepper, you'll have to get it yourself because I don't know how to get it online".
The planner can understand things like "there's risotto rice, jasmine rice, arborio rice, etc.; some recipes ask for specific type of rice but some recipes are OK with any kind of rice". And there's more things I plan to support in the future, like ingredients that allow "use this ingredient or this other ingredient", or optional ingredients.
When we get a plan, we then check we have all the needed stock, and if we don't, we update the stock and replan. And when it's done, it generates an ICS file with calendar entries we add to our shared calendar ("today we're cooking gnocchi with pesto for 4 people, here's link to the recipe"). It also generates stickers, which you print on sticker paper, and you put them on the ingredients that will be needed to cook. The stickers ensure someone doesn't accidentally eat too much of them before we need them.
It's given us a lot of utility for our household. It lets us make tasty fancy meals without the toil of planning. It's quite rough around the edges - no fancy UI, lots of "you have to enter this textproto here and there's next to no validation". But I'm slowly improving it in my spare time to make it more convenient and useful for us.
Right now, the published part of it is not useful, it only includes like 8 recipes that we manually entered. But I'm trying to design it so that it can eventually be a platform, and ideally easy-ish to use for households.