SK Yugoria introduced a corporate ModelOps platform on Wednesday

2024: ModelOps Platform Implementation

The Ugoria Group of Insurance Companies introduced the corporate ModelOps platform on Wednesday. The project was implemented on open technologies by GlowByte, which announced this on February 20, 2024. The created space is a single platform that brings together employees involved in the development and implementation of ML models throughout the company. The solution made it possible to quickly create models, conduct their training and integrate them into workflows.

The platform meets the main requirement of the insurer - it provides ease of use of the tool for all categories of users, from developers to IT. This was achieved through the implementation of a development methodology, the organization of a working environment for the Data Science team, and the creation of a system that can easily scale to the needs of the business.

The main challenge is to make MLOps available both in terms of maintenance and development, while at the same time not depriving the solution of flexibility. Our goal was not just to install components, but to give users a convenient and understandable methodology that will solve more complex problems. The selected technology stack has effectively closed the customer's needs. In addition, these tools are relatively easy to maintain, "said Grigory Shutov, senior architect of Advanced Analytics GlowByte.

The platform is integrated data with the insurer's sources and consists of various cluster tools. Kubernetes

ML models are developed through JupiterLab, and with the help of the Gitlab CI/CD tool, a single pipeline of model output from the development stage to production is built. Airflow is used as an orchestrator for the use of ML models.

We managed to build a convenient and fast solution for MLOps, available not only to financial institutions, but also to customers who are just starting to build MLs in their infrastructure. This is another good example of building on Open source in the insurance sector, said Evgeny Chernoburov, head of insurance practice GlowByte.

The project prospects include the development and implementation of the backlog, improving QCD and integration by ML, building optimal processes. It is planned to increase the number of models and development teams, as well as the introduction of new tools and their customization to requirements.