DevOps / Production Engineer

Profile

A graduate information systems engineer from EPISEN, I spent two work-study years at Crédit Coopératif – Groupe BPCE developing, integrating and industrialising banking applications: Jenkins, Artifactory, XL Deploy, SonarQube, Checkmarx, automated tests, release preparation and run support. On top of that, three years of a team engineering project gave me hands-on practice with Linux, Docker, Kubernetes, Terraform, Ansible, AWS, on-premises private cloud, GitOps, observability and incident diagnosis.

Professional experience

Crédit Coopératif – Groupe BPCE · Paris · IT Studies and Data department, Java team

• Industrialised Jenkins CI/CD pipelines for Java / Angular projects: artifacts in Artifactory, automated tests, SonarQube and Checkmarx scans, deployments with XL Deploy.
• Prepared releases: prerequisite checks, change documentation and coordination of delivery steps.
• Supported the run: defect tracking, reproduction, first-level diagnosis and coordination with users, IT teams and production.
• Tracked tasks, dependencies, defects and milestones in Jira and Confluence, in Agile Scrum / TEMPO: daily stand-ups, planning, reviews and retrospectives.
• Built Angular interfaces and Java / Spring services, with REST APIs, data persistence and error handling.

Highlights

SIRIUS engineering project

EPISEN · 2023 – 2026 · team of 3

• Move from Jenkins to GitHub Actions and Argo CD with GitOps: Image Updater, Argo Rollouts, Blue-Green and Canary strategies.
• Security and observability: Keycloak and OAuth2 / OIDC, RBAC, NetworkPolicies, Prometheus, Loki, Grafana, probes and resource limits.
• Several Kubernetes clusters, separate dev, staging and production environments, per-environment configuration with Kustomize.
• Hybrid infrastructure: about 10 AWS VMs provisioned with Terraform, VMware vSphere VMs at the school, Ansible configuration and a Proxmox lab with snapshots and rollback.

Engineering thesis

• Controlled protocol: 10 Kubernetes incident scenarios, 15 repetitions, 2 models and 3 architectures (LLM only, RAG, ReAct agent), for 900 trials blind-scored by 2 LLM judges.
• With Kimi K2.5, the ReAct agent reaches 5.113/6 versus 4.230/6 for classic RAG, at the cost of about 77.84 s of extra latency and about 34,258 extra tokens.
• Recommendation: classic RAG by default (4.120/6 with Qwen in about 4.04 s), agentic escalation for complex incidents. FastAPI service tested, containerised and deployed with GitHub Actions and Argo CD.

Education

Engagement

• Maths tutor with Acadomia since 2023: 8 middle and high school students supported with tailored explanations, targeted exercises and progress tracking.
• I chaired the chess section of Crédit Coopératif’s sports association (ASCC) and the EPISEN chess club: running the clubs, organising activities and bringing members together. I handed both presidencies over after graduating. Competitive chess player, notably during my preparatory years in Cameroon.