Otkritie Bank launches technological platform for machine learning models

2021: Launch of ML Model Development and Implementation Platform

On July 8, 2021, Otkritie Bank announced the beginning of pilot operation of a technological platform for the development and implementation of machine learning models. Its architecture is recognized as bank-wide. This modeling platform will reduce operational risk and allow you to quickly implement models of any complexity into business processes. The technology partner of the project is Neoflex.

The implemented project will ensure the continuity of the process of developing and implementing machine learning models and their subsequent integration into. business processes bank The deployed architecture automates the management of models, work with them in a separate safe test environment, and automatically transfer changes to the industrial environment.

Otkritie Bank managed to build processes for the development of modeling teams, providing them with advanced technologies. To develop machine learning models, access to enterprise data warehouses is implemented, ready-made workplace images for teams with a different stack of machine learning libraries are configured.

Automated processes of transferring models from the development loop to the application loop will in the future reduce the time-to-market tasks of introducing machine learning into the bank's decision-making processes, as well as reduce the operational risk of manually transferring model logic between systems.

With the introduction of the industrial modeling platform, the issue of connecting a new specialist to the team of modelers should take no more than one day. Each development team works in its own environment with dynamic expansion of resources for model training. We also have the ability to quickly transfer models from the development environment to the application environment for various teams. It was due to such flexibility that the platform was chosen as the basis for the bank-wide platform, "commented Pavel Nikolaev, managing director of the integrated risks department of Otkritie Bank.

The platform will give bank modelers the opportunity to use current development methods and algorithms in a user-friendly environment, and will increase the time-2-market of models. The native and important goal of the platform is to reduce operational and model risk when introducing and operating models, "said Natalya Khozinskaya, Chief Data Scientist of Otkritie Bank.

To implement the modeling platform, the Neoflex team applied the MLOps methodology and practices in the bank - combining the technologies for developing ML (Dev) models and operating developed ML (Ops) models. Automation and monitoring of data, models, processes at all stages of operation of the modeling platform is a solid foundation for the further development of machine learning processes at Otkritie Bank, "said Gennady Volkov, partner, chief architect of Neoflex.

The simulation platform was implemented on the basis of an industrial environment deployed in the bank, containers Kubernetes which provided the platform with the necessary flexibility and scalability. The bank is working on the migration of data used in modeling to the industrial circuit, which should provide the platform with even greater stability and fault tolerance, "said Dmitry Pervukhin Otkritie Bank, vice president, director of the department for the development of accounting and analytical systems.

2020: Launch of the technology platform for machine learning models

Otkritie Bank on October 22, 2020 announced the launch of a technological platform for machine learning models.

Thanks to this platform, the bank will improve the monitoring and decision-making processes in the segment of legal entities and individual entrepreneurs through the use of multifactorial statistical models for assessing credit risk.

Otkritie Bank launched a project to automate the life cycle of machine learning models on the technological stack of open software: Jupyter Lab, Airflow, MLFlow, Jenkins, Minio. Bank experts, together with Neoflex, a technological partner of the project, prepared the infrastructure and launched the MLops platform in pilot mode. It allows you to manage the versioning of model scripts, work with them in a separate safe test environment, and automatically transfer changes to the industrial environment.

Within three months, the project team managed to solve architectural problems and implement this approach for one of the business-critical models. The vector of further development in Otkritie Bank of the infrastructure for the development and implementation of machine learning models, as well as big data analysis has been determined.

This pilot project will further automate the management of statistical models, solve problems related to the development of banking products based on automated decision-making, enrich monitoring and decision-making processes with data and analytics, "said Pavel Nikolaev, managing director of the integrated risk department of Otkritie Bank.

The created infrastructure will further create a continuous process of developing and implementing models and reduce the time for their integration into the Bank's business processes, "said Ekaterina Lazaricheva, director of the risk technology center of the Otkritie Bank's integrated risk department.