RecipeRec

Damijah Carter

Tempe, Arizona

2 0

AI-Enabled Shopping Cart Assistant that suggest recipes to shoppers as they shop and add items to their cart ...learn more

Project status: Concept

Artificial Intelligence

Intel Technologies
OpenVINO, DevCloud

Code Samples [1]

Overview / Usage

Grocery Stores in the United States generated 759.57 Billion USD in sales for 2020 (Coppola, 2021). RecipeRec is an AI-enabled shopping assistant aims to help grocery stores exceed this number. As customers shop and add products to their cart our RecipeRec suggest additional products customers can purchase to fulfill a suggested recipe based on the items already in their cart.

For customers, making a grocery list alleviates the number of trips back to the grocery store due to a forgotten ingredient, saving shoppers time and money but shoppers often lack a shopping list. According to SpendMeNot, 68% of women and 52% of men make a shopping list before going grocery shopping (SpendMeNot, 2021). RecipeRec rescues shoppers who did not make a list by recommending additional products customers can purchase to fulfill a recipe suggested by RecipeRec based on the items already in the cart. This is not only useful to customers by exposing them to new dishes and ensuring they are not forgetting any ingredients, but this cross-selling strategy can increase profits for grocers.

Methodology / Approach

RecipeRec uses an image detection model to identify items placed in the cart or scanned at checkout from a live video. The live video is then quantized in TensorFlow Lite to reduce the model size to improve CPU and hardware accelerator latency.

Next, Intel’s OpenVINO classification model is used to classify what objects exist in each frame generating a list of items in the cart. The list of items in the cart is displayed on a small screen where with the touch of button shoppers can request suggested recipes that require one or two additional ingredients/products to complete the recipe.

Once the button is activated the list of ingredients/products is sent to a script in python that hails the Spoonacular API. Where the APIs complex food ontology is leveraged to provide a list of recipes that require one or two additional ingredients to complete the recipe given a set of ingredients. These lists of recipes are displayed to the shopper as a guide to ensure they are not missing essential ingredients and to expose them to new dishes.

At checkout, a similar process takes place. The live video from checkout is quantized in TensorFlow Lite and Intel’s OpenVINO classifies the objects in each frame to generate a list of items purchased. Upon customers completing payment the Spoonacular API is again used to provide a list of recipes to shoppers to expose them to new recipes that they can make with one or two additional ingredients.