ENERGIES

Innovative architecture for AI

#Cloud Data Science

#Dataiku

#Energies

#Ambition

Ambition
of the mission

As a historical user of Dataiku, our client built a datalab in 2015 whose main tool is a Dataiku platform, on which a wide variety of AI projects have been developed, from the simplest use cases to the most advanced. The DSS platform initially deployed on the customer's historical datacenter suffered from significant technical debt, and no longer allowed new use cases (such as high-volume image analysis) to be addressed. It was therefore decided to redesign this platform to take advantage of the advantages of the cloud and gain in performance, scalability and flexibility.

#Method

Our
approach

Think

Reinventing a cloud platform for AI

As our client's uses are rapidly increasing in complexity, it became necessary to design a more flexible platform, capable in particular of analyzing several hundreds of millions of images daily, while maintaining maximum ease of use for more traditional use cases.

Make

Federating business requirements and technological constraints

A tripartite working group has been set up, reconciling the datalab and the IT teams. The objective: to meet the strong constraints of security and cloud IT practices, while selecting the AWS services most suited to the needs of the projects! The switch from Legacy environments to the new platform has been made as transparent as possible for users, thanks to the development of specific migration tools, guaranteeing the non-regression of the existing perimeter.

Scale

Guarantee scalability over the coming years

In order not to fall back into technical debt, our team designed a platform offering business users total autonomy in the deployment of execution environments (Spark clusters, GPU environments, etc.) while optimizing budgets via an on-demand approach. Without safeguards, this type of approach can introduce a risk of stability; it was therefore also necessary to think of an innovative deployment method, allowing an incremental hot backup and an automatic restoration of the environment in case of failure.

#Benefits

Indicators
of success

Substantial gains of computing power and execution capacity

High resilience of the platform, self-healing environment

Reconciliation of practices Datalab and IT

#DataStories

Discover our
other use cases

See all our use cases