Food regulatory compliance check platform

Project description

API service was created, which extracts product information from the image uploaded/taken by the user – that includes nutrition information, weight, and manufacturer. Inconsistent quality of the images and variety of product packaging makes this task quite challenging.

A dedicated team has trained custom machine learning models using Open Food Facts (open-source dataset) and around 100k product images from Amazon. Data was labeled with the help of our 3rd party partners. In combination with several cloud OCR services, we managed to deliver high-quality API service.

The pipeline with several steps was created to process a big amount of data in a scalable manner.

Technologies:

• Customized YOLO implementation on PyTorch.
• spaCy + sci-kit-learn for NLP adjustment.
• Dockers and Google Cloud for continuous delivery.
• Google OCR and AWS Textract.
• Scrapy to collect data from different sources.