← All work

Delivery engineering · Platform & infrastructure

GitOps delivery for a payments-critical platform

a London-headquartered DeFi protocol · 2023–present

Focus
Delivery engineering · Platform & infrastructure
Evidence
3 documented proof points
Decision record
Firsthand trade-off included

The brief

What was built

Rebuilding release engineering around ArgoCD and Helm so that rolling, blue-green and canary deploys became routine rather than events.

Evidence

Recorded outcomes

  • Median pipeline 35 → 20 minutes, measured as ArgoCD sync plus Helm rollout, 90-day medians either side
  • Environment provisioning from days to under an hour via multi-account Terraform and Ansible
  • 99.9% monthly availability SLO with an enforced error-budget policy

Decision record

One serialized commit queue, and no cancellation

The first GitOps workflow failed for an embarrassingly concrete reason: it wrote to `config/rc-appset`, while the real manifests lived under `config/bank-appset`. It also allowed an in-progress deployment update to be cancelled by a newer run, which is unsafe for a repository acting as deployment authority. I replaced it with one serialized GitOps commit queue — cancellation disabled, environment validated, no-op updates made explicit, rebase before push. After that, every release became a commit with a reproducible image SHA, and rollback became a Git revert that ArgoCD could reconcile.

Return to all work