this post was submitted on 25 Dec 2023
1928 points (97.9% liked)
People Twitter
4854 readers
1872 users here now
People tweeting stuff. We allow tweets from anyone.
RULES:
- Mark NSFW content.
- No doxxing people.
- Must be a tweet or similar
- No bullying.
- Be excellent to each other.
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Cameras in your fridge and pantry to keep tabs on what you have, computer vision to take inventory, clustering to figure out which goods can be interchanged with which, language modeling applied to a web crawler to identify the best deals, and then some conventional code to aggregate the results into a shopping list
Unless you're assuming that you're gonna be supplied APIs to all the grocery stores which have an incentive to prevent this sort of thing from happening, and also assuming that the end user is willing, able, and reliable enough to scan every barcode of everything they buy
This app basically depends on all the best ai we already have except for image generation
Rolling this out for tools and parts at my work. Tool boxes with cameras in the drawers to make sure you put it back. Vending machines for parts with auto order.
Cameras and computer vision aren't necessary. Food products already come with upcs. All you need is a barcode reader to input stuff and to track what you use in meals. Tracking what you use could also be used for meal planning.
Yeah, I did think of the barcode approach, but I didn't think anyone would be willing to scan every item, which is why I ignored it
However, revisiting this question made me realize that we could probably have the user scan receipts. It would take some doing but you could probably extract all the information from the receipt because it's in a fairly predictable format, and it's far less onerous.
OTOH, you still have to scan barcodes every time you cook with something, and you'd probably want some sort of mechanism to track partial consumption and leftovers, though a minimum viable product could work without that
The tough part, then, is scouring the internet for deals. Should be doable though.
Might try to slap something together tonight or tomorrow for that first bit, seems pretty easy, I bet you've got open source libraries for handling barcodes, and scanning receipts can probably just be done with existing OCR tech, error correction using minimum edit distance, and a few if statements to figure out which is the quantity and which is the item. That is, if my adhd doesn't cause me to forget
If you can also keep recipes in the system you could skip scanning the barcodes here. You'd just need to input how many servings you prepared and any waste. Even if the "recipe" is just "hot pocket" or something. If the system knows how much is in a package it can deduct what you use from the total and add it to the list when you need more.
Tracking what you use would be a lot easier with AI. Then you wouldn’t have to keep a barcode scanner in the kitchen. You could just have a camera pointed at your food prep space
is AI good enough to manage that with just a camera? how would it determine how much of a given product you uses? Like if you dump a cup of flour in a bowel, how does it know how much that was.
If you have to point the product in front of the camera to register it anyway, might as well use a barcode reader anyway because it's the same thing at that point just without the risk of the AI misidentifying something.
I think you can achieve a similar result by having one giant DB so we can average out general consumption and then have a personal/family profile, where we in the first place manually feed the AI with data like, what did we bought, exp date, when did we partly or fully consume it. Although intensive at first I think AI will increasingly become more accurate whereby you will need to input less and less data as the data will be comming from both you and the rest of the users. The only thing that still needs to be input is "did you replace it ?"
This way we don't need cameras
Oh, so you're saying that the only data the algorithm needs in the limit is whether or not the user deviated from the generated shopping list, and if so, how, right?
This is true, it's just a bit difficult to cross the gap from here to there
I’m stoked that you’ve got it all hammered out, let me know when you’re done.
Sure no problem, I just need you to puch in some data manually so we can get started. Can you get thid stack done by tomorrow? Awesome, see you tomorrow!