Platform / DevOps Engineer

Collaborated with the London team to pilot the infrastructure of DeepPCB, a reinforcement-learning product for autonomous printed circuit board routing developed with Google Cloud and NVIDIA.

• Managed a shared fleet of approximately 170 on-premise H100 GPUs with team-based quotas and multiple GKE clusters.
• Supported 20 to 30 ML, SRE, and platform engineers on a multi-million-pound annual cloud footprint, while mentoring a junior platform engineer.
• Designed a Crossplane operator in Python and Jinja2 enabling each team to provision cloud resources from a single claim with integrated tenant isolation.
• Implemented dual-level GitOps (Flux and ArgoCD on the GKE Connect gateway) to prevent Kubernetes API exposure.
• Migrated DeepPCB to the new platform, reducing provisioning time from several weeks to half a day.
• Rewarded the blue/green upgrade of RayService to prevent loss of in-flight work during model deployments.
• Configured automatic rollback and cancellation of Argo Rollouts on Prometheus analysis above 5% errors or 5s p99.
• Migrated DeepPCB inference from Argo Workflows to KubeRay to package the model as an independent FastAPI endpoint.
• Served 6 client organizations simultaneously on the same hardware with 200 to 500 inference tasks per day per client.
• Configured DCGM with Prometheus and Grafana to alert on allocated but idle GPUs or ECC/XID errors.
• Deployed Kueue with gang scheduling on AWS EKS with the NVIDIA GPU operator and MIG partitioning.
• Migrated ingress from nginx to Envoy Gateway for native traffic splitting of the Gateway API.
• Led the migration of the LangChain Cooper agent backend to internally hosted H100 bare-metal servers with vLLM.
• Led the consolidation of Neptune to MLflow for over 10 AI engineering teams.
• Reduced annual GCP costs by approximately one third while building a per-client cost dashboard.
• Technical environment: GCP, AWS, Kubernetes, GKE, EKS, Crossplane, ArgoCD, Flux CD, Argo Rollouts, KubeRay, Ray Serve, vLLM, Python, Jinja2, Prometheus, Grafana, DCGM, Kueue, KEDA, Vault, Envoy Gateway, MLflow.

Built the company's entire GCP footprint from scratch using Terraform, transitioning the structure from a strictly on-premise domain to infrastructure as code.

• Overhauled CI/CD to move production releases from weekly cadence to multiple daily deployments with fast rollback.
• Operated the production Kubernetes platform for a multi-tenant travel SaaS booking engine grouping approximately 20 microservices on GKE.
• Managed cluster upgrades, node pool sizing, HPA tuning, and on-call rotations.
• Deployed services using Helm via ArgoCD.
• Installed and operated self-hosted Kubernetes on OVH bare-metal servers using Ansible.
• Ran Istio on both clusters with a sidecar per service managing traffic routing and mTLS.
• Administered Proxmox virtualization and pfSense site-to-site VPN.
• Deployed Prometheus, Grafana, Loki, and Jaeger for centralized observability and drafted the disaster recovery plan.
• Technical environment: GCP, GKE, Terraform, Kubernetes, Helm, ArgoCD, OVH, Ansible, Istio, Proxmox, pfSense, Prometheus, Grafana, Loki, Jaeger.

Migrated client workloads from on-premise to AWS as a consultant.

• Built the target infrastructure as code with Terraform and executed it on EKS, including cluster autoscaling and IAM roles for service accounts (IRSA).
• Rebuilt CI/CD for backend microservices and mobile applications with shared pipeline stages.
• Integrated SAST and DAST analysis behind quality gates.
• Technical environment: AWS, EKS, Terraform, IAM, IRSA, CI/CD, SAST, DAST.