Most of the Insurance, Banking and even industrial companies hold a huge amount of data. Extracting the maximum of value is the real challenge and any treatment requires to be compliant with the privacy and the associated regulation. On top of this, Machine and Deep Learning treatments are currently one of the most promising technology giving a competitive advantage by extracting more intelligence from the data. However, these technologies are complex to integrate, requiring skilled personnel and IT flexibility. In this anxiety context, this conference illustrates how the use of Big Data solutions, together with a Deep Learning platform abstraction called Max-ICS can make this easy and flexible. We will illustrate this by creating a data treatment pipeline that will use deep learning on a very dynamic and ephemeral infrastructure that will use a private farm, an extension to public clouds, an extension to on-premises servers and finally using the nightly unused capabilities of office workstation to improve the algorithm efficiency.