Engagements run from architecture framing to production rollout, under one rule: deliver a
system your team can operate and evolve without me.
01 — Cloud
Design the infrastructure before building it
A serious technical design document saves months of rework. It settles the network, the
sizing, the backup strategy, the disaster recovery plan and the compliance requirements
before the first line of code.
Hub-and-spoke topology: cross-cutting services — egress gateway, firewall, DNS, logging, bastion — are pooled once instead of duplicated per environment.
Architectures meeting French ANSSI requirements, SecNumCloud qualification
studies, government private cloud, secure remote administration and network segmentation.
Migration and recovery
Migration paths, disaster recovery plans that are designed and actually
tested, gradual cutover rather than big-bang.
AWSAzureGCPOVHOpenStackNutanixVMwareKubeVirt
02 — Kubernetes
Share without mixing
The question is never whether you need Kubernetes, but how many clusters and how to
partition them. A well-designed shared cluster costs less and runs better than ten
dedicated ones.
Isolation comes from admission policies and quotas, not from multiplying clusters. Each tenant gets its own space while the platform layer stays common.
Clusters on demand
Automated provisioning across the full lifecycle: creation, deletion, node
pools, address ranges, service accounts. Driven by API and by workflows.
AKSGKERKE 1 & 2OpenShiftKarbon
Multi-tenancy
Isolation through Capsule and hierarchical namespaces, quotas, network
policies, permission management per profile — read, develop, operate.
CapsuleOPA GatekeeperKyvernoRBAC
Purpose-built operators
Operators written in Go with Kubebuilder or in Ansible with the Operator
Framework, to encode your business rules inside the cluster itself.
KubebuilderGoOperator Framework
Cost control
Vertical and horizontal autoscaling, smart eviction, spot instances.
Capacity tracks actual demand.
KarpenterKEDAcast.ai
03 — Solution architecture
Decide before coding
A solution architect turns a business intent into a system that holds — and documents why
one option won over another.
Review of the existing estate, technical debt identification, a costed
and prioritised path out of it.
Modelling
UML analysis and modelling, OpenAPI 3, ISTQB and ISO 9126 standards,
code quality metrics and coding rules.
04 — Automation
Whatever repeats should run on its own
Infrastructure as code is a discipline rather than a tool: the Git repository becomes the
single source of truth, and production stops being edited by hand.
A complete GitOps pipeline. Quality gates — tests, static analysis, vulnerability scanning, policies, load — block a release before it ever reaches production.
Infrastructure as code
Terraform modules, Ansible roles and playbooks validated with Molecule,
images built with Packer, declarative management through Crossplane.
TerraformAnsibleMoleculePacker
GitOps
Continuous reconciliation through ArgoCD or FluxCD, generic Helm charts,
environment promotion by merge request.
ArgoCDFluxCDHelm
Observability
Prometheus, Grafana, Alertmanager, exporters written from scratch when
nothing off the shelf fits. Dashboards and alert rules come documented.
PrometheusGrafanaDatadogFluent Bit
05 — Artificial intelligence
AI as an engineering tool, not an argument
Two concrete, proven uses: accelerating design and development, and orchestrating business
processes through supervised agents.
AI-accelerated development
Daily pair programming with Claude Code. The gain lies less in typing speed
than in the ability to explore several architectural options, generate the tests and keep
documentation current.
Agent-driven automation
n8n workflows orchestrating specialised agents — analysis, execution,
review — wired into business systems and infrastructure, with full traceability and human
approval on sensitive actions.
