GlowByte introduces corporate ModelOps platform in Auchan

2023: ModelOps Platform Completed

On March 9, 2023, GlowByte announced that it had completed a project to implement a corporate ModelOps platform on open technologies in the AUCHAN Retail Russia retail network. The solution allowed Data Science specialists to create an environment for the design and commissioning of ML models.

Speed and accuracy are a prerequisite for the company's competitiveness in the retail market. Together with GlowByte, we implemented a project to create a unified environment for working with big data, which allows you to manage ML models written in any language, - commented Maxim Strozhny, Chief information officer of AUCHAN Retail Russia. - Thanks to the project, we have improved the efficiency of Big Data by implementing predictive and recommendation models based on machine learning. I note the flexibility and reliability of the platform - we were quickly able to rebuild and adapt to work on new software.

When building the platform architecture, the teams were guided by the principles of the MLOps approach to the development of ML models. It is a set of practices and technologies that combine Machine Learning, DevOps, Data Engineering, and Model Governance into a single methodology for creating, implementing, and operating machine learning models.

The source platform integrated is data retailer deployed VK Cloud in and consists of various cluster tools. Kubernetes This allows for tool flexibility, efficient management of available computing resources, and rapid scaling.

The system for storing persistent data (GlusterFS + Heketi) is deployed on the domestic operating system Astra Linux. ML models are developed in Python using JupiterLab, and using the Gitlab CI/CD tool, a single pipeline of model output from the development stage to production is built.

Data Science experts manage the life cycle of ML models through an Open Source platform based on MLFlow. It allows you to conduct various experiments - to log the metrics and parameters of the model, make decisions about its implementation, perform a retrospective analysis of the process of changing metrics and models. Airflow is used as an orchestrator for the use of ML models.

"Thanks to the MLOps platform implemented in the AUCHAN Retail Russia trading network, all ML models are now developed on the basis of a single template and have a standardized productivity pipeline for routine prediction and automatic retraining," said Alexander Kukhtinov, head of practice at ModelOps GlowByte. - Now the process of outputting Python models has been developed on the platform, but in general, the flexibility of the tool allows you to work with any models, including R Java,,. C/C++ Test runs showed a decrease in both time-to-market and time to maintain and retrain the model, which means that data scientists and IT subdivisions will have the opportunity to solve more problems. "