emetralGet in touch

Emile Metral · France & Mauritius

Reliable platforms.Intelligent operations.

I design multi-cloud and Kubernetes platforms that stay up, and the AI agents that make them easier to run.

Get in touchAbout me

01 — About

Four years keeping business‑critical systems up.Now bringing AI agents into how they're built and run.

I'm a Solution Architect for AI & Cloud Infrastructure at TeamWork, in Mauritius, designing AI‑driven cloud solutions for clients across Europe and Africa.

Before that I was an SRE / Platform Engineer at Veolia in Lyon, France: production GKE, Kubernetes security, a hybrid network spanning three clouds and on‑prem, low‑downtime migrations and multi‑cloud FinOps. Production first, guardrails always, humans in the loop.

4+
years running business‑critical platforms in regulated environments
100+
servers operated across a hybrid GCP · AWS · Azure · VMware estate
3
clouds in production, plus on‑prem — one consistent way of running them

Certified

  • Certified Kubernetes AdministratorCNCF · The Linux Foundation
  • Certified Kubernetes Security SpecialistCNCF · The Linux Foundation
  • Professional Cloud Security EngineerGoogle Cloud
  • Solutions Architect — AssociateAmazon Web Services

02 — Work

Six areas, one way of working: built to run in production.

01Platform EngineeringExplore
02Migrations & Hybrid NetworkingExplore
03Security & ObservabilityExplore
04Applied AI & AgentsExplore
05FinOps & Cloud CostExplore
06Built for MyselfExplore

Cloud, Kubernetes and Linux — down to the kernel. Multi-cloud foundations, shared Kubernetes platforms and the Linux servers underneath — run, tuned and debugged so product teams never have to think about nodes, keys, pipelines or I/O.

Moving critical systems without the 3 a.m. rollback. Replatforming and migrations planned for low downtime, on top of a hybrid network that ties on‑prem, GCP, Azure and AWS together.

Secure by default. Seen before it breaks. Guardrails from the admission controller to the endpoint, and observability that answers real questions during an incident.

Agents that do real work, with guardrails. Generative AI wired into real workflows — phone lines, chat, ticketing, image pipelines, operations — with tests, fallbacks and humans in the loop.

Every euro accounted for, across three clouds. Cost visibility and automation across GCP, AWS and Azure — so spend is explained, allocated and switched off when nobody is using it.

Side projects, run like production. Things I build because I want them to exist — and run with the same standards as client work.

Contact

Have a platform to make reliable,or an agent to put in production?

[email protected]

01 — Area of expertise

Platform Engineering

Cloud, Kubernetes and Linux — down to the kernel.

Multi-cloud foundations, shared Kubernetes platforms and the Linux servers underneath — run, tuned and debugged so product teams never have to think about nodes, keys, pipelines or I/O.

  1. GCP first, plus AWS, Azure, Scaleway and on-prem VMware/Nutanix in production — one consistent operating model.
  2. Shared multi-tenant GKE: private control plane, Workload Identity everywhere, node auto-provisioning, Gateway API.
  3. Everything as code: Terraform/OpenTofu, Helm/Helmfile, CDK and CloudFormation; Puppet and Ansible for the fleet.
  4. Self-managed GitLab end to end, with CI runners on Kubernetes that scale to zero overnight.
  5. Linux in depth: server fleets, network-stack and storage analysis, production troubleshooting — and the runbooks and mentoring to share it.
  6. Performance engineering: Go profiling for >100 % gains, K6 load tests with custom Go extensions.

Selected work

  • Veolia

    Shared GKE platform

    Multi-tenant clusters, Gateway API, node auto-provisioning, Spot opt-in.

  • Veolia

    CI runners on Kubernetes

    Pod and ephemeral-VM executors, warm pool by day, zero by night.

  • TeamWork

    AWS multi-account estate

    Accounts, guardrails and self-service via Service Catalog.

  • Skyloud

    EKS & Kapsule platforms

    Terraform-built clusters, Helm charts, ArgoCD GitOps.

Full interactive presentation in progress.

Tools

  • GCP
  • AWS
  • Azure
  • Scaleway
  • Linux
  • Docker
  • GKE
  • EKS
  • Kapsule
  • Terraform
  • OpenTofu
  • Helm
  • CDK
  • GitLab CI
  • ArgoCD
  • Puppet
  • Ansible
  • Go
  • Python
  • Bash
  • PowerShell

02 — Area of expertise

Migrations & Hybrid Networking

Moving critical systems without the 3 a.m. rollback.

Replatforming and migrations planned for low downtime, on top of a hybrid network that ties on‑prem, GCP, Azure and AWS together.

  1. VMware vSphere 5.5/6 → 8 upgrades with custom low-downtime processes.
  2. Fleet replatforming from CentOS to AlmaLinux and Debian.
  3. Hybrid network with an on-prem hub: IPsec, Partner Interconnect, Azure ExpressRoute and HA-VPN with BGP across clouds.
  4. Network services run end to end: Cisco ASA, pfSense and iptables firewalls, Bind DNS, load balancing, Squid proxies.
  5. On-prem to AWS with Application Migration Service, and a hands-on VMware to Nutanix move.

Selected work

  • Veolia

    Datacenter replatforming

    Hypervisor and OS generations upgraded without stopping the business.

  • Veolia

    Hybrid cloud network

    15+ inter-site links, routing drift checked against Terraform intent.

  • TeamWork

    Cloud migrations

    Critical on-prem workloads to AWS; VMware to Nutanix.

Full interactive presentation in progress.

Tools

  • vSphere
  • Nutanix
  • AlmaLinux
  • Debian
  • BGP
  • IPsec
  • HA-VPN
  • ExpressRoute
  • Partner Interconnect
  • Shared VPC
  • AWS MGN
  • OpenVPN
  • Cisco ASA
  • pfSense
  • iptables
  • Bind
  • Squid

03 — Area of expertise

Security & Observability

Secure by default. Seen before it breaks.

Guardrails from the admission controller to the endpoint, and observability that answers real questions during an incident.

  1. Identity across every cloud: GCP IAM, AWS IAM, Azure Entra ID and Keycloak — least privilege, workload identity, no static keys.
  2. Security and access audits that end in a prioritised remediation plan, inside an ISO 27001 scope.
  3. Kubernetes security as code: Kyverno admission, Tetragon runtime enforcement, Cilium network policies and mTLS.
  4. Endpoint protection at scale: Defender for Endpoint for 1,200+ users, EDR alerts turned into tickets.
  5. Observability built for incidents: Prometheus, Grafana, Loki, OpenTelemetry, SLOs and on-call.

Selected work

  • Veolia

    Policy guardrails

    Admission and runtime policies that developers can actually debug.

  • TeamWork

    Endpoint security rollout

    1,200+ users, EDR alerts to ITSM tickets via a severity matrix.

  • Veolia · Skyloud

    Observability stacks

    Prometheus, Grafana, Loki, Mimir — from scratch and at scale.

Full interactive presentation in progress.

Tools

  • GCP IAM
  • AWS IAM
  • Entra ID
  • Keycloak
  • Kyverno
  • Tetragon
  • Cilium
  • Vault
  • CrowdStrike
  • Defender
  • Prometheus
  • Mimir
  • Grafana
  • Loki
  • OpenTelemetry
  • Alertmanager
  • Opsgenie

04 — Area of expertise

Applied AI & Agents

Agents that do real work, with guardrails.

Generative AI wired into real workflows — phone lines, chat, ticketing, image pipelines, operations — with tests, fallbacks and humans in the loop.

  1. Voice and Teams IT help-desk agent that qualifies requests, files ITSM tickets and escalates only when needed.
  2. Image-compliance pipeline driven from Slack: an agent orchestrates large-scale classification with human review.
  3. Hybrid AI inference: on-prem open models with a cloud burst path, for data that has to stay home.
  4. AI for operations: agent-ready knowledge bases, MCP access to metrics, a Claude Code platform for engineers.

Selected work

  • TeamWork

    AI help desk

    Amazon Connect, Bedrock, MCP tools, end-to-end conversational tests. In production.

  • TeamWork

    Photo compliance

    Strands agent, Step Functions at scale, Nova Lite with guardrails, content-hash dedupe.

  • TeamWork

    Hybrid inference

    Greengrass + Ollama on-prem, SageMaker HyperPod and Bedrock in the cloud.

  • Veolia

    Agent-ready knowledge base

    A wiki in Open Knowledge Format, one concept per page, built for LLMs.

Full interactive presentation in progress.

Tools

  • Amazon Bedrock
  • AgentCore
  • Strands
  • Amazon Connect
  • Q in Connect
  • MCP
  • RAG
  • Vertex AI
  • Gemini Enterprise
  • Claude
  • AWS CDK

05 — Area of expertise

FinOps & Cloud Cost

Every euro accounted for, across three clouds.

Cost visibility and automation across GCP, AWS and Azure — so spend is explained, allocated and switched off when nobody is using it.

  1. Multi-cloud cost pipeline: AWS usage data federated into BigQuery, one set of dashboards for GCP, AWS and Azure.
  2. Kubernetes cost allocation per namespace, Spot capacity and scale-to-zero.
  3. Scheduled stop/start of non-production environments, packaged as a self-service catalog.

Selected work

  • Veolia

    Multi-cloud cost reporting

    Keyless AWS → GCP transfer, BigQuery, Looker dashboards.

  • TeamWork

    Off-hours scheduler

    Tag-driven EC2 / RDS / Aurora stop-start with dependencies and snooze.

Full interactive presentation in progress.

Tools

  • BigQuery
  • Looker
  • AWS CUR
  • Step Functions
  • EventBridge
  • Service Catalog

06 — Area of expertise

Built for Myself

Side projects, run like production.

Things I build because I want them to exist — and run with the same standards as client work.

  1. A spaced-repetition app that your own AI assistant fills through an MCP connector.
  2. Native Mac, iPhone and Apple Watch widgets for Claude Code usage limits.
  3. Kubernetes-as-a-Service for edge machines, with hosted control planes.
  4. A self-hosted platform with no inbound ports, sandboxed containers and automatic rollback.

Selected work

  • Live

    Réviser

    Spaced repetition filled by your AI via MCP. Svelte, FastAPI, PWA.

    revision.emetral.fr
  • Daily use

    Claude Usage

    Liquid Glass widgets on Mac, iPhone and Watch. Swift, WidgetKit.

  • In progress

    OneKonsole

    Edge Kubernetes-as-a-Service: Kamaji, Cluster API, Go.

  • Runs this site

    Home platform

    Cloudflare Tunnel → Traefik → gVisor containers, all in Terraform.

Full interactive presentation in progress.

Tools

  • Svelte
  • FastAPI
  • Swift
  • Go
  • Rust
  • Kamaji
  • Cluster API
  • Terraform
  • Cloudflare