dimadem/upf-checker

This repository contains code for a project developed during the Encode hackathon, focusing on improving awareness of eating habits.

UPF Checker

See the demo here!

👋 This is UPF Checker - the AI companion on a mission to help you identify and reduce your consumption of Ultra-Processed Foods.

• The UPF Checker is a progressive web application that helps the user to identify the ultra-processed food and reduce its consumption by transforming the user selfie inducing the emotional response.

• Under the hood, the UPF Checker uses three models: an open-source Optical Character Recognition model that extracts ingredients as text, ChatGPT that identifies the food category based on the extracted text using the zero-shot learning, and Stability AI image-to-image model that transforms user selfie using the conditional parameters based on the identified category.

Why it matters

Ultra-processed foods (UPF) are products created through manufacturing processes that predigest raw food ingredients. These foods are quick to digest, often leading to overconsumption of calories and weight gain. From ready-to-eat meals to packaged snacks and sodas, UPFs are often low in nutrients but high in sugar, salt, unhealthy fats, and additives. Recent studies have linked UPFs to a host of health issues, including obesity, cardiovascular diseases, and even some cancers.

UPFs account for a significant portion of the total dietary energy consumed in many countries. In high-income nations like the UK, Canada, and the USA, they make up more than half of the total dietary intake. Even in middle-income countries such as Brazil, Mexico, and Chile, ultra-processed foods contribute between one-fifth and one-third of the total dietary energy.

However, until now, there has been no simple method to identify these products easily.

The UPF Checker addresses this issue by providing a straightforward way for users to recognize ultra-processed foods. It aims to help policymakers prevent potential health crises by raising awareness about the harmful effects of ultra-processed foods. The UPF Checker achieves this through an engaging and entertaining approach, making the identification process more accessible and appealing to users.

Project structure

• Research folder contains scripts with model training experiments we conducted and used to select our best final models that can be found in the backend folder.
• Our backend is built using the Flask Python web framework. It has a sigle endpoint for our client to post requests.
• Our frontend is a progressive web application built using the Next JS. It utilises the IndexedDB to store our user photo locally on the user device. Tailwind is used for the design.

Meet the team

👩‍🏫 Emeli Dral - Machine Learning expert and DevOps

👨‍💻 Dima Dem - Frontend Developer and Designer

👨‍💻 Bakhtiiar Bakhautdinov - Backend Developer

👩‍💻 Kate Usacova - Fullstack Developer and Solution Architect

Try it out!