Mediascope with support from Neoflex launches new platform for development and implementation of machine learning models

2022: Launch of the platform for the development and implementation of machine learning models

The research company Mediascope has launched a platform for the development and implementation of machine learning models into the industrial operation of Data Science. The company announced this on February 8, 2022. This platform will reduce operational risks and reduce the time-to-market integration of models into Mediascope business processes. The technology partner of the project was Neoflex.

As a result of the project, the company received a scalable and manageable space for developing ML models, which allows you to quickly connect internal teams of data scientists with the ability to evaluate the results of their work. With the help of the platform, the company will also be able to quickly and with minimal labor to attract external ML teams to increase the number of tasks and models being developed. In addition, a centralized catalog of finished pipelines will become available to specialists with the simplification of subsequent model development through the re-use of finished components.

As of February 2022, the platform has configured MLOps processes (model versioning, experimental tracking, assembly of executable services based on developed models) with the ability to track the origin of artifacts. The platform architecture provides an automated process for developing and implementing models, transferring them to the industrial environment, and providing tools for visualization experimental metrics. This allows you to reduce the development time, achieve reproducibility of results and increase the reliability of complex processing pipelines, data the elements of which are ML services.

To build the platform, Kubeflow open source solution was chosen, providing centralized tools for the development of ML models, pipelines and artifact management. In addition, Argo Workflow is used as the most developed orchestrator of workflows on Kubernetes, which is part of Kubeflow and facilitates the use of developed models.

The company has a streamlined process and its own tools for deploying machine learning models as services and incorporating them into data processing pipelines, however, in order to increase scaling capabilities, transparency of the process and reduce the time to put research algorithms into commercial operation, it was decided to develop a new DS platform. This makes it possible both for closer integration of internal teams, and if necessary, for the prompt connection of external teams to the development of new models with automated validation of the quality of the proposed solutions, "said Vasily Kuzmin, CIO Mediascope.

Mediascope is a technology company that uses a large number of complex ML models to process data and obtain analytics. Therefore, it was especially important for us to develop a solution that fully meets the high technological standards of our customer, and at the same time it would be convenient for data scientists to use daily. The introduction of this Data Science platform will allow Mediascope to reduce time-to-market for new analytical products based on machine learning models, as well as reduce the labor costs of validation teams and putting models into industrial operation, "said Alexey Antonov, associate partner, head of the Data Science competence center of Neoflex.