RATP
DIANE group data platform: architecture lead for the business squads, from framing to production, in a data mesh setup.
A single foundation for the group's reporting, big data, data visualisation and predictive models.
Enterprise
A group data platform, with business teams standing on their own
A data platform is not an infrastructure project: it is a foundation that ten business teams will use at once, each with their own use cases and pace. Without a shared architecture frame, every squad rebuilds its own plumbing and the platform becomes ten platforms.
Public transportLogistics and transport
Sound familiar?
If two of these sentences get said in your organisation, the problem is no longer technical: it is organisational.
- Every department has built its own warehouse, and nobody knows which one feeds what.
- A new use case takes months to get from prototype to production.
- Business teams wait on the data team for the smallest extract.
- Jobs fail overnight and nobody notices until the next morning.
How we do it
We set the architecture frame before the plumbing
Architecture documents, a technical consistency frame, patterns shared across squads. That is what stops the same problem being solved five different ways.
We make business teams autonomous without letting go
Self-service and a data mesh approach, yes, but with onboarding for each squad and support on their technical choices. Autonomy without support produces debt.
We industrialise the route to production
Continuous deployment, unit and regression tests, go or no-go checkpoints before release. The handover from build to run stops being the moment everything breaks.
We monitor, and we hand over
Monitoring of critical flows, documentation kept current, internal teams trained. If the platform only holds while we are there, the job is not done.
Where we have already done it
Each line is a real engagement, with what was measured at the end. Where there is no figure, nothing was measured: we do not invent one.
Geodis
Migration of processing from SAP to Cloudera and then Databricks, with the scheduling layer rebuilt on Airflow.
Spark jobs optimised in production and a scheduler switch with no service interruption.
The other subjects
Knowing what data you hold, and who may see it
A living catalogue, classified data, and access rules that survive an audit.
See the subjectWhen nobody believes the reporting figures any more
Reports nobody opens, an obsolete platform, and teams quietly going back to Excel.
See the subjectChanging ERP without losing the data on the way
Go-live never fails because of the software: it fails because of the data poured into it.
See the subjectBetween the database and the business, a layer is missing
The technical foundation is delivered, and the end user still has nothing they can use.
See the subjectDetecting what must be detected, without drowning the team
A regulatory alert engine running every day, and analysts working on cases, not noise.
See the subjectTell us where it is stuck.
Thirty minutes to lay out the subject and tell you frankly whether it is one for us. You leave with an opinion, whether you sign or not.
Book a slot30 minutes · no commitment