I develop, deploy and harden mission-critical platforms.

Designed and built the department's Cloud platform, from infrastructure as code to day-to-day operations: an internal offering for staff, a turnkey hosting foundation for the incubated government startups, and AI components deployed on a sovereign LLM.

• Infrastructure as Code — private networks, bastion, load balancers, managed databases and Kubernetes clusters defined in Terraform; configuration through idempotent Ansible playbooks, with guardrails and check mode before every run.
• CI/CD forge — self-hosted GitLab, autoscaled Kubernetes runners (20 parallel jobs), Harbor image registry, Traefik ingress and TLS certificates automated with cert-manager.
• Security by design — no server directly exposed: admin access through a filtered bastion, databases with no public endpoint, encrypted secrets (Ansible Vault), per-project robot accounts, automatic security patching.
• Resilience — encrypted off-site backups (GPG AES-256), snapshots before every risky operation, live migrations (registry across clusters, PostgreSQL 16 → 17) with zero data loss.
• Sovereign hosting — applications deployed from pipelines to the Clever Cloud PaaS (SecNumCloud), with Rancher and Metabase for operations and reporting.

Ran and maintained an estate of 10 applications and POCs: GitOps standardisation of the deployment chain and hardening of the software supply chain, from commit to production.

• GitOps — each environment's state is versioned in Git and reconciled by ArgoCD on OpenShift: traceable deployments, reviewed in merge requests, rollback in a single commit.
• Standardised pipelines — a shared GitLab CI chain (build, tests, Helm packaging, promotion across environments): every new project inherits a complete chain from its first commit.
• Secure supply chain — Trivy vulnerability scans of images on every build (shift-left), artefacts and dependencies served from controlled registries (Nexus), no deployment outside the pipeline.
• Observability — Prometheus metrics, Grafana dashboards and alerting to catch incidents before users do.

Built aidejuridictionnelle.justice.fr from scratch: the national service where citizens apply for legal aid online and staff process their applications.

• Architecture — Angular front end and Java Spring Boot REST API, with asynchronous processing decoupled through ActiveMQ messaging to handle nationwide volumes.
• Two audiences, one product — from the citizen's online application to its processing by staff, on sensitive personal and financial data.
• Industrialised — containerised application (Docker), shipped through Jenkins pipelines to OpenShift and monitored in production with Prometheus and Grafana.
• Quality first — code reviews and automated tests on a critical public service, where every regression affects citizens.

Centralised management of prison incidents — inmates and facilities — in a single application covering the whole French prison estate.

• Full stack — Angular front end and Java Spring Boot / PostgreSQL REST API: reporting, tracking and consolidating incidents in a single system of record.
• Shipping under constraints — automated deployments (Jenkins, Docker) on a heavily constrained legacy infrastructure: reliable, repeatable releases.
• Sensitive context — confidential prison data, with security and traceability built into the design.

Document-generation POC built with software vendor OpenText to produce the Ministry's business-application documents: document templates and A3 PDFs.

• Adopted by projects — used in French Ministry of Justice projects, notably to build document templates.
• Critical applications — documents generated from business data and consumed by critical business applications.
• Full stack — Java Spring Boot / PostgreSQL REST API and Angular interface, shipped through Jenkins pipelines.

Inmate-transfer management application, giving a consolidated view of how the prison population is distributed across facilities.

• High-visibility delivery — run in Scrum, shipped in short sprints in a tight loop with the business.
• Industrialised stack — containerised Symfony and PostgreSQL (Docker), continuous integration and deployment with GitLab CI/CD.

Online Peugeot and Citroën configurators — a key step in the buying journey — deployed in every country where the two brands are sold, totalling 3.2 million vehicles sold in 2017.

• Front & back — PHP Symfony / PostgreSQL back end and responsive React interfaces: a rich configuration experience (model, trim, engine, colour, options, price) for the general public.
• One platform, every market — a single product rolled out to every country where the two brands are sold.
• Method — delivered first in a V-model, then in Scrum: hands-on experience of both delivery cultures.

Development and evolution of the association's website (back-office, content, user journeys) for the start-up DonkeyCode.

• General-assembly voting — rebuilt a feature where reliability and result integrity are non-negotiable.
• End to end — features shipped from the Symfony / MySQL back-office to the Twig templates, at start-up pace.

Real-time billing-control tools for Orange Business Services' large international accounts (WS-Publication, Manual Billing 50D).

• Financial stakes — real-time monitoring of billing flows to catch anomalies early on major international accounts' invoices.
• Software quality — PHP Symfony / Doctrine code backed by PHPUnit unit tests, versioned with Git.
• Distributed team — daily work, in English, with developer and QA teams based in India.

Information-systems expertise supporting the customer service team: handling complex cases, then automating bespoke dashboards — needs analysis, extraction of raw data from the IS, cross-referencing, consolidation and formatting. Final tools presented to customer service managers.