Ten years in infrastructure, eight of them running Kubernetes in production. 50 clusters, 1,000+ nodes, from bare metal to multi-cloud. What I leave behind: a platform your team runs without me.
The whole technical foundation, hardware to developer self-service: a private OpenStack cloud, 13 Kubernetes clusters and 250 nodes, 72 RabbitMQ clusters and 127 PostgreSQL databases. 110 applications moved onto it with no interruption on business-critical scope.
Cluster build 1 hour → 5 minutes. Incident qualification time down 80% with an on-premise LLM agent.
Multi-cloud AWS and Alibaba platform: a majority of workloads on ECS/Fargate plus 6 Kubernetes clusters, about 200 nodes and 500+ microservices under continuous deployment for the group's houses, sized to absorb private sales and Fashion Week. Co-owned the worldwide customer and loyalty database — Couchbase, 2 TB, XDCR across three regions.
Capacity matched to real load with Karpenter and KEDA, not provisioned for peaks.
The shared internal GitLab platform deployed on Kubernetes for 2,500 developers and 350 projects, with its runner fleet industrialised on Kubernetes and reusable Terraform modules adopted across the ministry.
Job queue time down 70%. Environment provisioning 1 day → 2 hours.
Short public-cloud engagements alongside the long missions: audits of architecture, reliability, security posture and cost, delivered as a prioritised remediation plan — and the builds that follow. One GCP estate rebuilt from scratch as code after a security breach.
Most of the public-cloud work happened on these two. The rest of the stack is below.
EKS and ECS under continuous deployment for 500+ microservices, five Terraform / Terragrunt / Terramate foundation stacks across accounts, Karpenter and KEDA for elasticity, IAM and IRSA for access, with RDS, Aurora, ElastiCache, SNS, SQS and CloudWatch underneath. Crossplane self-service on top, so a product team ships application and infrastructure in one merge request.
40 microservices off Elastic Beanstalk in a year. Delivery pipeline 20 minutes → 3.
Managed Kubernetes on GKE and event pipelines on Pub/Sub for independent clients, run alongside the AWS work — including one client's entire estate rebuilt from nothing as infrastructure as code after a security breach: assessment, reconstruction, hardened access and network exposure.
A compromised estate back in production, entirely as code.
Tools are evidence, not the argument. This is what is in production today.
Kubernetes — EKS, ACK, GKE, RKE2, Talos, Kubespray, OpenShift, Magnum · Docker · Helm · Kustomize · Karpenter · KEDA · Kyverno · Crossplane
AWS · OpenStack · GCP · Alibaba Cloud · OVHcloud · Scaleway · Cloudflare · Pure Storage · Linux (Debian, Ubuntu, RHEL)
Terraform · OpenTofu · Terragrunt · Terramate · Spacelift · Ansible · Helm · OpenStack HEAT
GitLab CI · GitHub Actions · Azure DevOps · Argo CD · Flux · Renovate · Jenkins · blue/green deployment
Prometheus · Grafana · Loki · Mimir · Tempo · Alloy · Faro · OpenTelemetry · Alertmanager · Datadog · Sentry · Fluent Bit
vLLM · KServe · HolmesGPT · Kagent · LiteLLM · Milvus · RagFlow · MCP · AWS Bedrock · Hugging Face
HashiCorp Vault · Boundary · Consul · Keycloak (OIDC) · SOPS · Trivy · Falco · IAM/IRSA · policy-as-code
Couchbase · PostgreSQL (CNPG) · Cassandra · MySQL · MongoDB Atlas · Redis / Valkey · RabbitMQ · ClickHouse · Ceph · MinIO
SLI / SLO / error budgets · DORA metrics · blameless post-mortems · on-call · toil reduction · developer self-service · FinOps · DevSecOps
Paris, fully remote. Seed-stage through regulated enterprise. Contract or permanent.