Web dev student here and I'm interested in knowing more about creating products that actually use AI to help its users (not products that just use GPT in the backend). More specifically, I want to build a food supply management app for restaurants for my school thesis. This app will use AI to analyse food supplies and assign them purchase priority, value, and complexity scores (maybe just priority if it's too hard). Restaurant owners could then determine what foods should be purchased before others based on the priority scores.
For example, a restaurant may only have 10 tomatoes left and the average usage of tomatoes in this restaurant is 12 per week. Based on this, a priority would be assigned to purchase x amount of tomatoes.
Other factors that could be taken into account for the priority score could be:
- The difference between the average price that the restaurant bought tomatoes at and the current market price (a lower difference means that complexity is lower).
- The availability of tomatoes in the area (retrieved from a static survey or by calculating the amount of tomato sellers in the town)
- The current day of the week (higher priority as weekends get closer, i.e. more clients come on weekends and less tomato sellers are available)
- The amount of tomatoes to be bought (it's harder to buy 100 tomatoes than 10)
- The amount of time/effort it took to purchase tomatoes based on previous historical data
So, given some inputs like the above mentioned, how can I get the app to calculate priority scores for each food in a restaurant? Also, how could the app host multiple restaurants as users? I believe that this would mean that each restaurant has individual AI models (not every restaurant may evaluate each factor in the same way or with the same priority).
Over the top of my head, I believe that I should store a restaurant's historical data of previous food orders (including the scores) and, for example once a day, I train an AI model (linear regression model?) with that data and produce a score with the current food data. So in a way, it's using AI to generate scores on-demand without storing the actual AI model in a database (not sure if this is possible). Would this be the recommended way to do this?
PS: I want to clarify that this project idea is purely for educational purposes. I understand that people may think that this is a horrible product idea, it's too complex, or that restaurant owners don't need this or won't use this, but my intention is not to sell this product or provide it as a service but to learn how to actually build AI products.
[link] [comments]