Gabin CHAMBON

B2B telecom operator. Moved into a dedicated AI role: owner of the group's generative AI platform — GPU foundation, LLM serving, model gateway, agent fleet. No public egress, ANSSI-PA-102 / NIS2 scope. • Initiated, designed and shipped the generative AI and agent platform: vLLM serving driven by KubeAI, KV-cache-aware routing (Gateway API Inference Extension), LiteLLM gateway — per-team virtual keys, token budgets, audit log. • Deployed the NVIDIA RTX PRO 6000 Blackwell GPU foundation: VFIO passthrough via Nova, strict NUMA affinity, drivers as Talos system extensions, DCGM monitoring. Model placement without NVLink, FP8/NVFP4 quantisation, KV cache sizing. • Deployed and run a fleet of 20+ agents (Hermes, Kagent): alert analysis, CVE and ticket triage, cluster checks, Kubernetes sandboxes, Teams app — remediation through merge requests only. 90 % of tickets handled within the hour, incident qualification −80 %. • Built agent and model evaluation (golden dataset, blocking CI smoke tests, LLM-as-judge, shadow runs: 13 regressions blocked) and agent observability (Langfuse, Grafana): token cost −40 %. • Built the MCP servers and infrastructure CLI the agents use: typed tools, schema-constrained arguments, read-only identities. • Integrated coding agents into the IDEs (FIM completion, TTFT p95 < 250 ms; repository agents) behind a single OpenAI-compatible endpoint. • Industrialised the model lifecycle: catalogue in Git, signed weights (cosign) verified by Kyverno, blocking promptfoo evaluation in CI, Langfuse tracing, Argo Rollouts canary promotion. RAG on pgvector/CNPG, OKF knowledge base, n8n workflows. vLLM · KubeAI · LiteLLM · Gateway API Inference Extension · MCP · Hermes · Kagent · Langfuse · promptfoo · GuideLLM · pgvector · n8n · NVIDIA GPU · DCGM · Talos · OpenStack · cosign · Kyverno · Python · Go

B2B telecom operator. 7-person SRE team owning the whole foundation — OpenStack across 3 datacenters, servers, storage, Kubernetes, app delivery. 99% target. • Operate the whole foundation, hardware to application: OpenStack across 3 datacenters, 13 stretched Kubernetes clusters, 250 nodes, servers and Pure Storage arrays. • Designed the platform Helm chart behind the group's 200 applications and run the data layer through Operators: 72 RabbitMQ, 127 PostgreSQL (CNPG), 60 Redis/Valkey, a 3 TB / 18-node Cassandra, MySQL. • Migrated 110 internal applications and servers onto the private cloud, no interruption on critical scope. • Initiated and shipped the on-premise LLM platform, GPU foundation to agents — hardware, OpenStack, Kubernetes/Talos, vLLM and KServe behind the Gateway API Inference Extension, document RAG, HolmesGPT. Signed model weights, read-only IAM, ANSSI-PA-102 / NIS2. Toil automation, coding assistance, alert triage. • Migrated all 13 clusters from Kubespray to Talos: build time 1 hour → 5 minutes. • Wrote and rehearsed the DR plan: etcd restore, full cluster rebuild, CNPG / RabbitMQ / Pure Storage backups, RTO and RPO per scope, validated by restore drills. • On-call and end-to-end incident management: triage, crisis calls, remediation, blameless post-mortems — incident qualification down ~80%. • Defined the SLIs/SLOs and the observability behind them (Prometheus, Mimir, Loki, Tempo, Grafana Alloy, Faro). • Standardised GitOps (Argo CD), policy-as-code (Kyverno), KEDA autoscaling on RabbitMQ, Boundary access, Keycloak/OIDC identity via Terraform. Kubernetes · Talos · Terraform · OpenStack · Helm · Kustomize · Argo CD · vLLM · KServe · GAIE · HolmesGPT · Kagent · Hermes · Milvus · Ansible · Kyverno · KEDA · Kubernetes Operators · cosign · HashiCorp Boundary · Keycloak · NVIDIA GPU · DCGM · Prometheus · Alertmanager · Mimir · Loki · Tempo · PostgreSQL/CNPG · Cassandra · RabbitMQ · Redis/Valkey · GitLab CI · llm-d · OpenTofu · Kolla · Bifrost

Global luxury group. 10-person infrastructure team. Multi-cloud AWS / Alibaba platform: 500+ microservices under continuous deployment — mostly on Amazon ECS / Fargate, the rest across 6 Kubernetes clusters (3 EKS, 3 Alibaba ACK, ~200 nodes). Crossplane self-service, ~50 product teams. 99.9% target. • Ran both of the group's deployment planes: the majority Amazon ECS / Fargate estate, administered in Terraform (services, task definitions, autoscaling, ALB), and the multi-cloud Kubernetes fleet. • Co-owned, with one other engineer, production support for the group's worldwide customer and loyalty database on Couchbase / Amazon EKS: 2 TB, 18 nodes, Magma storage engine, XDCR across 3 regions, run through the Couchbase Autonomous Operator. • Took part in the group's crisis calls on major incidents — diagnosis, cross-team coordination, remediation, post-mortem — on a platform sized for private sales, product drops and Fashion Week peaks. • Owned observability for that scope — Couchbase instrumentation, dashboards and alerting on Mimir / Loki / Grafana / Alertmanager — to catch replication and capacity drift before customer impact. • Tuned elasticity across both planes — ECS autoscaling, Karpenter and KEDA on Kubernetes (SNS and RabbitMQ queues, ingress and system metrics) — to match capacity to real load. • Industrialised the AWS and Alibaba estate in Terraform, Terragrunt and Terramate across 5 foundation stacks (base, IAM, network, databases, compute), rebuilding the dependency tree. • Maintained the shared deployment building blocks — Terraform ECS modules and the Kubernetes Helm chart — and their GitHub Actions pipelines; Kyverno policies, IAM/IRSA, HashiCorp Vault secrets. AWS (ECS/Fargate, EKS, RDS, Aurora, ElastiCache, SNS, IAM/IRSA) · Alibaba Cloud ACK · Kubernetes · Karpenter · KEDA · Crossplane · Terraform · Terragrunt · Terramate · Helm · Argo CD · Kyverno · HashiCorp Vault · Kubernetes Operators · Couchbase · Mimir · Loki · Grafana · GitHub Actions

3 days/week over three years, run alongside one other engagement at a time. Shared internal GitLab platform for central government IT — 5-person product team (3 DevOps, 1 tech lead, 1 product owner) serving 2,500 developers across 350 projects. • Deployed and operated the Ministry's GitLab CE platform for 2,500 developers across 350 projects — first on OpenStack with Ansible and NGINX, then migrated onto Kubernetes. • Industrialised the Ministry's GitLab runner fleet on Kubernetes (RKE2, OpenStack Magnum, Helm, Argo CD) behind a shared Squid egress proxy: job queue time down 70%. • Published 8 reusable OpenStack Terraform modules adopted across the Ministry's DevOps community — full environment provisioning from 1 day to 2 hours. • Migrated infrastructure as code from OpenStack HEAT to Terraform, and relocated the whole platform to a new government OpenStack tenancy with no service interruption. • Built the observability stack (Prometheus, Grafana, Loki, Mimir, Fluent Bit) and DevSecOps CI/CD pipelines for Terraform and Ansible; administered HashiCorp Vault on a Consul backend. OpenStack · OpenStack Magnum · OpenStack HEAT · RKE2 · Kubernetes · Terraform · Ansible · Helm · Argo CD · GitLab CE · GitLab CI · HashiCorp Vault · Consul · Prometheus · Grafana · Loki · Mimir · NGINX · Squid · Linux · Bash

2 days/week, run in parallel with the Ministry engagement (3 days/week). Staffing group, 20-person digital division. Objective: exit Elastic Beanstalk and containerise the application estate. • Migrated 40 microservices off Elastic Beanstalk within a year, choosing service by service between Amazon ECS for simple workloads and Kubernetes for the more complex ones. • Cut continuous delivery pipeline duration from 20 minutes to 3 (−85%), while improving application performance and observability coverage. • Onboarded the development teams onto Docker — reference images, local development stacks, training and hands-on support — through to full team autonomy. • Hardened the cloud estate (IAM, network), implemented CloudWatch alerting, and administered Amazon RDS and MongoDB Atlas. AWS (ECS, EC2, Elastic Beanstalk, RDS, IAM, CloudWatch) · Kubernetes · Docker · Terraform · Ansible · GitLab CI · GitHub Actions · MongoDB Atlas · Linux

2 days/week, run in parallel with the Ministry engagement (3 days/week). 5-person infrastructure team, historically focused on systems and network administration. Objective: move off a Proxmox / LXC estate onto Kubernetes at OVHcloud. • Migrated 50 applications from LXC containers on Proxmox to Kubernetes on OVHcloud — Helm charts, GitLab continuous delivery pipelines, SOPS-managed secrets. • Introduced blue/green deployment for those 50 applications through DNS CNAME switching, on infrastructure that previously took the service down at every release: zero downtime, instant rollback, and lead time for changes cut from 30 minutes to 5. • Deployed the LGTM observability stack (Loki, Grafana, Tempo, Mimir) alongside Prometheus and Datadog, and rolled coverage out to all 50 migrated applications — metrics, logs and traces from the moment each service landed on Kubernetes. • Deployed Consul Service Mesh to secure internal traffic. • Trained the team — systems and network administrators, not Kubernetes people — to deployment autonomy on Kubernetes, Helm, Kustomize and SOPS. Kubernetes · OVHcloud · Helm · Kustomize · SOPS · Consul Service Mesh · GitLab CI · Prometheus · Loki · Grafana · Tempo · Mimir · Datadog · Keycloak · Redis · NGINX · Linux

Founding technical hire — full-time until June 2021, then 2 days/week alongside the Ministry engagement. Architecture, automation and R&D on confidential computing in secure enclaves. • Designed end to end the OpenStack / RKE2 platform hosting the Geowallet application and Intel SGX enclave workloads (Scone), and owned it in production. • Migrated all AWS resources to OVHcloud OpenStack and automated provisioning through Terraform and Ansible (RKE2, CouchDB, PostgreSQL). • Built the GitLab CI/CD pipelines and Helm templates for the Node.js microservices application, with security embedded in the chain (Trivy, Falco, Snyk) and Prometheus / Grafana / Loki observability. • Developed a Go automation tool on the OVHcloud API; deployed Keycloak on Kubernetes and administered CouchDB and PostgreSQL in production. OpenStack · RKE2 · Kubernetes · Intel SGX / Scone · Terraform · Ansible · Helm · Docker · GitLab CI · Go · Node.js · CouchDB · PostgreSQL · Keycloak · Trivy · Falco · Snyk

Atawiz is an IT consulting firm with 20 employees, providing services to various companies such as Colas, Printemps, Sixens, Eramet, and Crédit Agricole. I joined the company as a DevOps Engineer, working within a team specializing in Cloud and DevOps. This team provided expertise across multiple client contexts. During my mission, I worked closely with the CTO on developing a Proof of Concept (POC) for the Hyperledger Fabric blockchain. My main responsibilities as a DevOps Engineer included: - Designing cloud architectures on AWS and Azure for various client accounts - Administering Kubernetes clusters (EKS, GKS) - Developing a POC for a decentralized application based on the Hyperledger Fabric solution - Deploying and administering Hyperledger Fabric on on-premise Kubernetes using Ansible and Helm - Writing Smart Contracts in Go for the Hyperledger Fabric blockchain - Creating and maintaining CI/CD pipelines (GitLab, Azure DevOps) for Node.js and PHP applications - Managing infrastructure with Terraform - Setting up monitoring stacks with Datadog, Prometheus/Grafana, and Loki Technical context: Kubernetes, AWS, Azure, Azure DevOps, Ansible, Terraform, Helm, Docker, Linux, Git, Bash, Golang, Grafana, Prometheus, Loki, Datadog, GitLab, GitHub, AKS, EKS, Hyperledger Fabric.

Equativ is a company specializing in Ad Tech, providing innovative and high-performance systems for online advertising. I joined the company as part of a work-study program, alternating one week of classes with two weeks in the company to complete my degree in IT Analysis. I worked as an IT Analyst within the Infrastructure team, which consisted of six System Administrators and DevOps Engineers. My main missions as an IT Analyst included: - Deploying and administering Kubernetes clusters using Kubespray (on-premises, KVM). - Writing and maintaining Terraform modules on AWS. - Creating and maintaining Ansible playbooks for Kubernetes and Linux administration. - Administering AWS resources (EC2, S3, IAM, Route 53). - Managing KVM virtual machines (CentOS 7). Technical context: Kubernetes, Terraform, Ansible, AWS, Linux, KVM.

Baobab is a company specializing in microfinance, with operations in several African countries as well as China. I joined the Information Systems team at the headquarters, which included a System and Network Administrator and a Cloud Specialist, as an IT Technician. My main missions as an IT Technician included: - System administration (CentOS 7, VMware). - Network administration (Ubiquiti). - Level 1 and 2 support for Windows and Linux. Technical context: Linux, Windows, VMware, Ubiquiti.