GlowByte and SberStrakhovanie reduced the deployment time of ML models in the insurer's system by 3 times

2023: MLOps Platform Implementation

SK SberInsurance together with IT the partner GlowByte implemented the MLOps platform, which allows you to regulate and standardize the process of deploying - ML models in the industrial environment. insurer The solution accelerated the deployment of models by 3 times and increased the efficiency of the department by Data Science systematizing the process of developing models and implementing new tools. This was TAdviser announced on November 10, 2023 by representatives of GlowByte.

The system allows you to download and process data, develop ML models, test them and implement them into the industrial circuit using CI/CD processes. The solution includes a number of functional modules and areas, including tools for analyzing and logging data, a model execution environment, as well as a repository for non-text and/or large files and a repository for model code.

The platform greatly facilitates and accelerates the work of Data Science specialists, Enabling you to select and run a development environment with a pre-installed set of libraries and data accesses work in the usual environment, create machine learning projects and manage their life cycle, log the results of experiments and track their metrics and metadata using convenient tools, package the ML model into a microservice and bring it into the industrial operation loop in the form of a docker container.

The launch of the project made it possible to implement a single system for carrying out a wide range of functions necessary to create and put machine learning models into productive execution. It also made it easier for developers, analysts and systems engineers to interact.

The purpose of this project was to automate routine processes. The platform we have implemented has a complex architecture, but the chosen technology stack has effectively solved the problem, despite the large number of integrations with data storage and processing systems. The project has already helped increase the number of models and development teams. We plan to improve the MLOps platform, creating more and more high-quality and effective solutions for SberInsurance, said Alexander Efimov, executive director of Advanced Analytics GlowByte.

In the future, the project is planned to implement a backlog, improve QCD and integration from ML, form the Feature Store, as well as increase the number of models and development teams.