GridWorks builds and operates the Kubernetes platforms, delivery pipelines and reliability practice underneath product teams — so deploys stop being an event and incidents stop being a surprise.
$ scanning 412 workloads · 3 clusters · 1,180 nodes ✓ control plane 1.29.4 3 AZ healthy ✓ argocd sync in-sync 0 drift ! requests/limits 64 workloads unbounded ✗ pdb coverage 11 deployments, no PDB ! idle spend $41,300/mo reclaimable ── 18 findings · 6 critical · report → gw-audit-0412.md $ availability 99.981% target 99.950% budget 62% left latency p99 45ms was 340ms -86.8% deploys 30d 1,842 rollbacks 3 MTTR 7m 12s $
Every engagement lands in the same place: infrastructure defined in code, delivery automated end to end, and an SLO you can actually hold someone to. Pick the entry point — the rest follows.
Production clusters with multi-AZ control planes, hardened node pools, sane admission policy and a tenancy model that survives your third team onboarding. EKS, GKE, AKS or bare metal.
Pipelines that build once, sign the artifact, and promote the same digest through every environment. GitOps reconciliation, progressive delivery, automatic rollback on SLO burn.
Error budgets with teeth: SLIs derived from real user journeys, burn-rate alerting that pages on symptoms, blameless post-incident review, and game days that break things on purpose.
Unit economics per service and per request, not a spreadsheet of EC2 line items. Rightsizing from live profiles, commitment strategy, and spend guardrails wired into the pipeline.
One OpenTelemetry pipeline for metrics, logs and traces with exemplar linking, so a latency spike takes three clicks to reach the offending span — and your ingest bill stays flat.
A golden path your engineers actually want to use: modular Terraform, Crossplane compositions for self-service databases and queues, and a paved-road template that takes a service to prod in a day.
# One module. Namespace, IAM, autoscaling, SLO and dashboards
# land together — or the plan fails.
module "checkout_api" {
source = "../../modules/service"
name = "checkout-api"
tier = "tier-0" # implies PDB + multi-AZ + page
runtime = {
replicas_min = 6
replicas_max = 120
cpu = "500m"
memory = "1Gi"
arch = "arm64" # graviton, ~22% cheaper
}
slo = {
availability = 99.95
latency_p99 = "120ms"
burn_alerts = ["14.4x/1h", "6x/6h"]
}
dependencies = [module.pg_checkout.endpoint]
}No click-ops, no snowflake clusters, no tribal runbook living in one engineer's head. If it can't be reviewed in a pull request and reconciled by a controller, we don't ship it.
Every engagement is scoped to an exit. We are done when your team is running the platform without us — and the documentation proves it.
Two weeks inside your infrastructure, pipelines and incident history. You get a findings report ranked by risk and cost, with a remediation plan that is costed in engineer-days.
We stand up the platform alongside your engineers: IaC modules, cluster baseline, delivery pipeline, observability stack. Everything in your repos, your cloud accounts, your name.
Workloads move behind a traffic-shifting cutover with a tested rollback at every step. We migrate the least critical service first and the payment path last.
Either we hold the pager under a signed SLA, or we run enablement until your team does. Most clients start with the former and move to the latter inside a year.
Follow-the-sun on-call across five timezones, an incident commander named before the page fires, and a public status timeline your account team can read without asking engineering. Every escalation ends in a written review — including the quiet ones.
The shape is deliberately boring. A signed artifact promoted by a GitOps controller into a multi-AZ cluster, with every layer defined in Terraform and every signal on one telemetry pipeline.
We have defaults because defaults are how you avoid a bespoke platform nobody can hire for. If you already run something that works, we keep it and document why.
Three engagements from the last eighteen months. Metrics are taken from the client's own dashboards, thirty days pre-cutover versus ninety days post.
01 Multi-region active/active
340ms p99 on the checkout path and a single-region failure domain. Rebuilt on Kubernetes with cell-based routing and asynchronous settlement.
02 Bin-packing the fleet
1,900 nodes running at 19% CPU utilisation. Bin-packing, Graviton migration, spot for the batch tier and storage lifecycle policy.
03 Trunk-based delivery
Hand-rolled Jenkins with a four-hour pipeline and manual QA gates. Replaced with trunk-based delivery, canary analysis and automated rollback.
Free, technical, and run by the engineer who would lead your engagement — not a salesperson. Bring your architecture diagram and your worst incident. You leave with a written summary of the top five risks either way.