Norilsk Nickel, together with GlowByte, has implemented a corporate ML platform based on the Kolmogorov product

2025: Implementation of an ML platform based on Kolmogorov software

Norilsk Nickel"" together with the integrator GlowByte implemented a vendor-based corporate ML platform Kolmogorov software Russian. Data Sapience The solution allows you to create a single environment algorithms machine learning with and implement a unified and scalable infrastructure for tracking the full lifecycle/ML AI applications and deep analysis. This was announced data by representatives of GlowByte on March 24, 2025.

The ML platform integrated with the corporate Data platform "" and Data Lake the system (DevSecOps corporate repository) and consists of various tools deployed in clusters, Kubernetes including the JupiterLab development environment , the MLFlow experiment management tool, the AirFlow batch process orchestrator, and the Grafana monitoring tools. ON Kolmogorov acts as a single entry point that connects these components into a whole ecosystem, improves the performance of all members of the Data Science team, and provides management of available computing resources and rapid scaling.

The platform architecture is designed with multi-user access in mind for a large number of specialists, including internal DS teams and external ML developers. 

The user interface for managing and productive work with the technology stack and open source tools is also implemented using Kolmogorov software. The tool allows you to launch and use machine learning model development environments, track the execution status of CI/CD pipelines of implementation, obtain information about the state of productive models, use prepared development project templates, provide controlled access to the registry of all projects for creating machine learning models that contain all the necessary information collected during the development and implementation of machine learning models. This comprehensive solution greatly simplifies both the work on ML projects and the coordination of the activities of individual development teams.

The Kubernetes environment allows you to scale applications and models to your current needs without major architectural changes, ensuring optimal resource utilization and cost savings. Automatic cluster and container management reduces maintenance and maintenance costs and improves application reliability and resilience. The modular approach provides the flexibility of the system and the ability to install additional components.

Thus, thanks to tools based on Kolmogorov software, DS teams have the opportunity to significantly reduce time-to-market and time-to-product indicators, as a result of which the company can respond faster to market requirements and quickly implement innovative solutions.

Since 2018, Nornickel has been consistently introducing artificial intelligence technologies into production and supporting business processes. The scope of application is growing, and this requires a new level of management of AI solutions, monitoring their performance and efficiency, improving the quality of work of model developers and Data Scientists. We solved this problem by combining data tools into a single ecosystem, part of which was the Kolmogorov solution. In general, according to the results of the project, we want not only to optimize existing processes, but also to create opportunities for business development through the use of data and machine learning, said Alexey Manikhin, Head of Business Application, Information Technology Department, Norilsk Nickel.

The project is based on the Kolmogorov platform, which combines popular open technologies for machine learning under a single intuitive interface. This allows specialists to focus on model development without being distracted by the technical difficulties of integrating various tools. This approach provides both the convenience of using ready-made solutions and the flexibility to customize for specific business tasks. The key quality of the platform is the balance between a ready-made industrial solution and the ability to adapt it to the individual needs of the customer, which is especially important for developing ML practices in large companies. noted Evgeny Lisitsin, Managing Director, GlowByte.

The Kolmogorov platform allows you to organize the management of MLOps components on-premium, making it easier for Data Science and DevOps specialists to perform tasks, as well as business roles. At Nornickel, the platform is responsible for the full cycle of managing ML solutions: model development, implementation and operation, while taking into account all the requirements of the customer's infrastructure, told Mikhail Zaytsev, owner of the Kolmogorov product, Data Sapience.

Further development of the initiative includes the creation of a separate, third circuit - Test landscape of the ML platform, integration into the corporate e2e process of MLOps, migration of existing machine learning models to the corporate ML platform with subsequent scaling. It is planned to transfer models of various types: Time series, Computer Vision, NLP and Large Language Models.