Uralsib, together with GlowByte, has developed functionality to create an MLOps platform

2025: Development of functionality for the creation of an MLOps platform

Uralsib, with the support of GlowByte, has developed the concept of a single approach for solving MLOps problems and the functionality for creating an MLOps platform. This will optimize computing resources for the use of machine learning models and ensure the stability of their operation in industrial operation. This approach has laid the foundation for effective management of the full life cycle of ML models, which opens up opportunities for the use of advanced tools for the development and operation of ML solutions. This was announced on June 4, 2025 by representatives of GlowByte.

The concept is based on the already established approach of MLOps, adapted by the GlowByte team to the needs of the bank. The technical implementation includes the Kubernetes cluster to manage computing resources and the JupiterLab web environment to develop ML models and experiments. To automate the implementation of machine learning models, a comprehensive Gitlab CI/CD system is used - using this tool, a single pipeline of model output from the development stage to use in industrial operation is built.

Within the framework of the project, a fairly new tool for orchestrating batch models, Prefect, was used for the market. The study allowed a comparative analysis with a generally recognized standard - Apache AirFlow. Based on the data obtained, it was possible to form an optimal architecture proposal for the bank to solve MLOps problems.

The creation of an MLOps platform is not just the introduction of technologies, but the formation of a new culture of working with data and models. Our team of MLOps specialists designed an architectural solution that allows you to standardize development processes, speed up the putting of models into commercial operation and optimize the use of computing resources. The project was especially valuable for the study of a new Prefect orchestration tool for the market, which made it possible to offer the bank an optimal technological stack, said Alexander Efimov, executive director of advanced analytics practice, GlowByte.

We consider information technology as the basis for business development and follow the path of active implementation of modern solutions. And if there are no ready-made solutions, they must be developed on their own. We are glad that we managed to implement the project together with GlowByte, a team of MLOps specialists whose team not only has powerful expertise in MLOps processes, but is also ready to participate in research on new components and offer a modern solution. noted Dmitry Fedorov, Director for the Development of Business Analytics and Machine Learning Platforms, Uralsib.

In the future, Uralsib plans to expand the architecture of the MLOps infrastructure with new components, as well as optimize the operation of DS teams by standardizing development approaches.