When I founded ACCENSEO in 2024, I set a non-negotiable rule from the very first customer: no manual configuration anywhere. Customer engagements touched healthcare, institutional real estate and finance, meaning production databases with several 100 GB of RAM (PostgreSQL, MongoDB), regular audits, and a need for full reproducibility across dev, staging, and production. Without IaC from day one, drift would set in within months.

I codified the entire infrastructure in Terraform: EC2 (application servers), RDS PostgreSQL (managed databases), S3 (object storage and backups), CloudFront (CDN), Lambda (serverless), API Gateway (REST exposure), EKS (container orchestration), VPC + Security Groups + IAM (network and security). Every customer environment has its own Terraform workspace with plans validated in GitHub Actions / Bitbucket Pipelines CI before any apply, Infracost plugged into the pipeline for explicit FinOps discipline (cost-delta review on every merge), and SSH tunnels for secured database access. Deployments are zero-downtime, backups automated, and disaster-recovery plans tested quarterly.

Zero manual configuration across the customer fleet, environments rebuildable in minutes on incident, and explicit FinOps present in every PR - any infra change displays its cost delta before being approved.

That discipline reshaped my commercial posture: I can promise a customer a reproducible environment and a transparent infra budget right from the quote, which sets me apart from consultants stacking ad-hoc servers. It is also the baseline I will replay in the next CTO scale-up role - treat infrastructure as a product deliverable, not as an ops chore.

The Pichet PSR platform (partner leads ingestion) ingested up to one lead every 2 seconds at peak, from a dozen external partners (SeLoger, Myopla, Cooper Advertising...) with strict SLAs. Each lost lead represented potentially tens of thousands of euros in missed real-estate revenue. Without per-partner observability we were flying blind - and an outage on a partner API could go unnoticed for hours.

I built observability partner by partner: dedicated SOFT Monitor dashboards (volume, error rate, latency) with one tab per connected API, real-time email alerts on every critical threshold, and native APIM observability (analytics, throttling, OAuth) on the Microsoft API Manager. I versioned the API across 5 consecutive documented versions on Confluence, with a progressive migration strategy for legacy partners. On infrastructure, I deployed on AWS EKS with Kubernetes + Docker + GitLab CI, and passed a formal 2023 security audit that hardened access controls and firewall rules.

Zero major lead-loss incident across 3 years, diagnostic time on cross-system anomalies reduced (from hours to minutes), SLAs respected on every partner, and partner integration lead time dropped from several weeks to a few days thanks to pipeline industrialisation.

That project locked in a reflex: invest in monitoring tooling on day 1 of a critical platform, because observability debt is never paid back at a reasonable cost. On every ACCENSEO engagement, that is the first deliverable I now lay down on any customer infrastructure I take over.

The Groupe Pichet ESB scope was more than 100 production integration flows across 20 business applications, 18K euros per month of Docker/Kubernetes hosting OPEX at Claranet, and 24/7 critical traffic on the accounting and financial flows. When I joined, every flow deployment relied on scattered manual operations and the SOFT Monitor system was firing 2,377 notifications per month with no triage capability.

I industrialised the pipeline brick by brick. On the CI/CD side, I rolled out a complete GitLab CI chain with explicit kill criteria on every deployment (tests, lint, Terraform plans). On operational quality, I imposed blameless post-mortems on every critical incident, formalised 7 types of technical documentation (DAA application architecture, DAT technical architecture, DAU automation, DEX exploitation, DFX flows, DIN installation, DMI migration), and one runbook per flow kept up to date. For observability, I ran the ELK Stack evaluation (Elasticsearch + Logstash + Kibana) to replace SOFT Monitor, and scoped the move to MongoDB Atlas for non-relational flows.

Single-digit incident rate maintained across 4 consecutive CIO changes (2021 to 2024), a fact often called out at the COPIL because it was unprecedented in the department. The post-mortem framework I shipped became the department-wide standard for every critical incident.

On this project I understood that DevOps maturity is not a tooling question but a discipline question: a simple system held in light SRE always beats a complex one abandoned after its purchase. That is the philosophy I lay down on every ACCENSEO engagement and that I will impose on the next scale-up platform.