Cloud-native data marketplace for SNCF

Context

For SNCF’s Data Market Place and the MyCot mobile application (Inventiv IT, 2025 – 2026, Paris) I designed a cloud-native data platform that enables large-scale data exposure and consumption. It is built as microservices on managed services, with integrated observability and monitoring, and I deployed the Spring Boot microservices on GCP that are used by 90,000 users.

Problem

A data marketplace serves two audiences with different needs: producers who expose data and consumers who use it, often through a mobile application. The platform has to scale with usage, stay operable by a small team, and show what it is doing, because a problem on a shared data platform affects every consumer at once. It also has to keep working as the number of producers, consumers and datasets grows, which favours small services that can be scaled and released independently.

Architecture decisions

A platform for exposing and consuming data at scale

The platform was designed as cloud-native from the start, to expose and consume data at large scale rather than as a collection of point-to-point integrations. Producers and consumers meet on the platform instead of on each other’s systems.

Trade-off: a platform approach costs more up front than a single integration. It pays back as the number of producers and consumers grows.

Microservices on managed services

The platform is made of Spring Boot microservices running on managed GCP services. Managed services shift operational work such as patching and scaling to the cloud provider, and microservices let each capability be deployed and scaled on its own.

Trade-off: managed services couple the design to one provider’s offering and pricing. In exchange the team spends its time on the platform’s behaviour instead of on running infrastructure.

Observability and monitoring built in

Observability and monitoring were integrated in the design rather than added after launch, so the platform’s behaviour under load is visible from the first release.

Trade-off: instrumentation adds development effort and telemetry cost. It is what makes incidents diagnosable on a system that many users depend on.

Outcome

The Spring Boot microservices deployed on GCP are used by 90,000 users, on a platform designed for large-scale data exposure and consumption and observable from day one.