Kubeflow-based machine learning platform for a pharmaceutical company

For a leading German pharmaceutical company with over 85 data scientists and 20 use cases, we built and operated a customised machine learning platform. The challenge was that until now there was no suitable ML development environment that supported the entire ML workflow from conception to implementation.

We set up a central platform based on Kubeflow, which was adapted to the company's strict regulations through customised functions. This was based on the "Kubeflow on AWS" distribution from AWS Labs, which provides key features such as Jupyter Notebooks, Pipelines and Serving. By implementing GitHub Actions and Terraform, we enabled efficient deployment of the development environment as well as the production environment.

To optimally support the entire life cycle of machine learning, we have integrated our own tools. These include MLFlow for tracking experiments, monitoring and logging. We also integrated multi-tenancy and profile management to fit seamlessly into the company's existing tool landscape. Single sign-on integration via Azure AD was also implemented to simplify access and ensure security.

The machine learning platform we provided enabled the pharmaceutical company to successfully operate and scale 20 different use cases. At the same time, costs were reduced by half. The centralised platform enables data scientists to collaborate seamlessly and run an efficient workflow. With support for GPU computations and automatic scaling, computationally intensive use cases were accelerated.

The implementation of the Kubeflow-based ML platform has enabled the company to cover the full range of the machine learning workflow. From the design and development of models to deployment and operation. This has enabled the data scientists to work more effectively and produce high-quality models faster.

Curious now? Let us show you what sets us apart from other companies and how we can help you achieve your goals.