Bringing a customer onto the SaaS platform meant creating their virtual machines by hand, one at a time. Terraform and PowerShell replaced the procedure: 3 to 5 hours saved for every new customer, depending on its size.
- Azure
- Terraform
- PowerShell
Mohamed Bekenniche
Infrastructure & cloud · 18 months in production
In a two-person team, I ran the production of a SaaS company: 12,000 users, four sites, two countries. I watched that production disappear on a Friday afternoon. Since then, I build infrastructure that can be destroyed and rebuilt, and I prove that it comes back.
Engineering student at Polytech · available for an internship from 18 May to 24 September 2027
mohamed@koveo: ~/portfolio
$ whoami --scope=production
role: Cloud & DevOps engineering student
experience: 18 months · SaaS company · team of 2
scope: 4 sites · 2 countries · 12,000 end users
migration: 10 applications → Kubernetes (AKS)
recovery: Velero · documented DR plan · tested restores
Background
2025 – 2028
Engineering Degree, Computer Science & Data, Polytech Annecy-Chambéry
Distributed systems, data and machine learning, governance.
Jun 2024 – Aug 2025
Infrastructure and Cloud Administrator (permanent contract), SYDEV
Kubernetes, CI/CD and GitOps, Terraform, observability, disaster recovery.
- 12,000 end users, with a two-person IT team
- 10 applications migrated to Kubernetes
- 3 to 5 h saved every time a new customer is onboarded
- 4 sites / 2 countries including Mauritius, interconnected by VPN
Feb – Jun 2024
Final-year internship, converted to a permanent contract, SYDEV
Terraform modules, account lifecycle automation with Ansible.
Jun – Aug 2023
Internship, systems and network support, WIDIP
Level 1 support for a multi-client managed services provider, documentation.
2021 – 2024
Bachelor of Technology, Networks & Telecoms, DevCloud track, IUT1, Grenoble
Skills
Skills followed by an arrow link to the case study or project that puts them into practice.
Build and deploy
Infrastructure described in a repository and delivered by a pipeline, rather than assembled by hand. This is where DevOps practice actually happens.
- Terraform
- OpenTofu
- Ansible
- Helm
- Azure DevOps
- ArgoCD (GitOps)
- GitLab CI
- Git
Run
The platform operated day to day: applications spread across several clusters, in two regions.
- Kubernetes (AKS)
- k3s
- Docker
- Azure Container Registry
- Harbor
- Traefik
- Linux
- Windows Server
- Azure
- OpenStack
- AWS
- GCP
Observe and recover
Knowing before the user does, and being able to roll back.
- Prometheus
- VictoriaMetrics
- VictoriaLogs
- Grafana
- Checkmk
- Velero
Connect and secure
Four sites, two countries, and access that has to stay bounded.
- IPsec VPN
- VLAN
- Routing
- DNS
- Multi-forest Active Directory
- Cryptography
- Privileged access management
Data
The databases the infrastructure carries (150 in production), and what they can be made to tell.
- MariaDB
- MySQL
- PostgreSQL
- MongoDB
- Microsoft SQL Server
- Data analysis
Automate and build tools
What replaces a manual procedure, and the small tools that make it usable by the team.
- Python
- PowerShell
- Bash
- REST APIs
- Streamlit
Support
Infrastructure serves people: helping them when something breaks, and leaving behind enough for someone else to take over without me.
- Technical support
- Remote assistance
- Technical documentation
- GLPI
Case studies
The hosting model no longer kept up with growth. Around ten applications were moved to AKS, spread across multi-region clusters by customer segment, and delivered through an Azure DevOps CI/CD pipeline and GitOps deployment with ArgoCD. The data stayed in the same managed pool: one database per customer, 150 in total.
- Azure
- Kubernetes (AKS)
- Helm
- Docker
- Azure DevOps
- ArgoCD
Cluster backups, a documented disaster recovery plan, restores that were actually tested. A backup that has never been restored is not a backup.
- Kubernetes
- Velero
- Azure
- Disaster recovery
Projects
Personal project
Backup tools restore only to the moments of their scheduled backups. BackReel records every change made through the Kubernetes API and journals every write to the volumes, to bring objects and data back as they were at any second. No access to etcd, and every restore states its real precision.
Proven
- 510 restores at random instants, not one acknowledged write lost
- the agent killed with -9 under load, taken over in under a second
- a disk rebuilt identical, byte for byte
- a namespace restored as it was, after a dry run
In progress
- the demo on a real cloud cluster
- Go
- Rust
- Kubernetes
- CSI
- ublk
- S3
- PostgreSQL
Personal project
A complete platform on the Infomaniak Public Cloud, built and destroyed with one command each: private network, three nodes across three zones, a k3s cluster, HTTPS ingress, GitOps with Argo CD, observability, and backups proven by a real restore. Every decision is recorded, and so is every failure. Next: measuring the energy and carbon of each workload.
Running
- private network, encrypted state, three nodes behind a bastion
- k3s cluster bootstrapped in a single apply
- HTTPS: Octavia, Traefik, cert-manager
- GitOps: Argo CD delivers the whole cluster
- observability: Prometheus and Grafana
- Velero backup, proven by a restore
- make up, make down, and CI on every change
In progress
- energy and carbon per workload, with their margin of error
- OpenStack
- OpenTofu
- k3s
- Argo CD
- Traefik
- Prometheus
- Velero
Articles
7 min read
What a shared Terraform state really costs
A command meant to delete a DNS record destroyed a SaaS company’s production infrastructure on a Friday afternoon. What follows is not an account of an operator error, but an analysis of what allowed an operator error to reach that far.
Read the article→
Contact
Available from 18 May to 24 September 2027
Assistant engineer internship: infrastructure, cloud, DevOps. Geneva (Switzerland), Annecy, Lyon, Grenoble, or remote.