01Platform · GitOpsAgicap · 2024 — NowKubernetes platform with GitOps CI/CD
49 applications and 8 core components run on the platform. Rollback time reduced from ~2h to under 90 seconds; 100% of infrastructure changes reviewed via pull request.
- →Built the platform on GKE with Terraform, ArgoCD, HashiCorp Vault, cert-manager, external-dns and keyless CI (Workload Identity Federation).
- →Moved deployments behind ArgoCD, so merging to main triggers the deployment automatically. Internal applications are packaged as Helm charts with separate values per environment.
- →Put Terraform behind Atlantis, so every infrastructure change goes through a pull request with the plan visible in the review.
- →Rebuilt the production cluster on a new IP range and cut over in one day (24/06/2026).
- →Set up Prometheus, Grafana and Loki for monitoring, with Slack alerts per team and synthetic probes.
< 90 sto roll back
100%infra by PR
49apps hosted
GKEAKSTerraformAtlantisArgoCDHelmVaultPrometheusGrafanaLoki
02Identity · Kubernetes operatorAgicap · 2024 — NowKubernetes operator that manages the whole Okta configuration from Git
2,871 Okta resources (745 accounts, 488 groups, 384 applications) are declared in Git and applied by the operator.
- →Built a Kubernetes operator that keeps Okta accounts, groups and applications in sync with the configuration stored in Git.
- →Added guards so a bad change cannot delete accounts or groups in bulk.
- →Changes are checked in a preview environment before they reach production.
- →12 alerts monitor the operator in production.
2,871resources in Git
745accounts
12alerts
KubernetesOperatorOktaGitOps
03Internal tools · LLMAgicap · 2024 — NowAlfred, an LLM IT-support bot on Slack
766 messages from 104 people in the first 30 days company-wide; only 12% became tickets.
- →Built a Slack bot that answers IT questions with Claude, grounded in a Notion knowledge base.
- →Runs on Kubernetes, on the internal platform.
- →Gives out local admin passwords directly; 72 were served in the first 30 days without a support ticket.
766messages
12%became tickets
72passwords, no ticket
ClaudeSlackNotionKubernetes
04Industrial · SEVESO · VoIPElkem Silicones · 2021 — 2023Telephony migration on a SEVESO high-threshold industrial site
Full VoIP migration completed for 570 employees with zero safety-related downtime.
- →Inventoried every phone, key line, and safety-critical station on-site before planning the migration.
- →Migrated in small batches, running old and new systems in parallel to allow immediate rollback.
- →Carried out the work outside production hours, with the operations team briefed and able to halt the migration at any point.
0 ssafety downtime
570migrated
100%inventoried up front
VoIPSIPLinuxVMware ESXIndustrial networkProject mgmt