As previously mentioned, a lot of work-related to data science happens on engineers’ laptops using Jupyter Notebooks. This creates a custom environment difficult (or impossible) to move into production.

In the case of data science platform services from cloud service providers such as Amazon Web Services® (AWS), Microsoft® Azure® or Google® Cloud Platform™ (GCP), these are predominantly “outsourced” Jupyter Notebooks with custom integration to their ecosystem (plugins). This allows them to deal with problems on a certain level, which is hard to replicate on-premises and on another cloud provider. Being cloud-agnostic and avoiding vendor locks are very important and a feature in the proposed data science platform solution using Kubeflow.