§ services / devops
Pipelines, Kubernetes, and on-call that holds.
Five capabilities, plus coverage we describe honestly. We make releases routine, systems observable, and incidents rare and short — then leave your team able to run it.
01
CI/CD
- Problem
- Releases are manual, risky, and rare, so fixes wait in a queue and Friday deploys are quietly forbidden.
- We do
- We build pipelines in GitHub Actions or GitLab CI with real gates — tests, linting, dependency and image scans — plus artifact versioning and safe rollouts, blue/green or canary, so a deploy becomes one boring button.
- You get
- A pipeline per service, a rollback path, and release notes generated from the change history.
02
Kubernetes
- Problem
- You are on Kubernetes, or should be, but it is fragile, over-permissioned, and nobody is quite sure what is running.
- We do
- We set up EKS with defaults that hold up — autoscaling, resource requests and limits, network policies, and Helm or Kustomize for deploys — and add the ingress, secrets, and RBAC that make it operable.
- You get
- A running cluster, deployment manifests in Git, and a runbook for scaling, upgrades, and the 2 a.m. calls.
03
Monitoring and observability
- Problem
- When something breaks you hear it from a customer first, and debugging comes down to guessing.
- We do
- We instrument metrics, logs, and traces with Prometheus, Grafana, and OpenTelemetry — or Datadog if you prefer one pane — define SLOs that match what users actually feel, and wire alerts that page on symptoms, not noise.
- You get
- Dashboards per service, SLOs with error budgets, and an alert set tuned to be worth waking up for.
04
Production support
- Problem
- Incidents are chaotic and undocumented, and the same failure keeps coming back.
- We do
- We stand up on-call rotations and escalation in PagerDuty or Opsgenie, write runbooks for the top failure modes, and run blameless postmortems that turn each incident into a concrete fix.
- You get
- A rotation, runbooks for the common incidents, and a postmortem trail that shows the failure rate coming down.
05
Automation
- Problem
- Engineers spend the week on manual toil — provisioning, backups, certificate renewals — instead of the product.
- We do
- We find the repetitive work and automate it: scripted runbooks, scheduled jobs, GitOps for deploys with Argo CD or Flux, and self-service tooling so a routine task does not need a ticket.
- You get
- Automated jobs with their own alerting, a GitOps setup, and a shorter list of things a human has to remember.
§ coverage
24×7, stated the way procurement can check it
We are a small senior team, not a staffed operations centre. Coverage means a defined on-call rotation with escalation and the response targets below — the kind you can put in a contract and hold us to.
| Tier | Scope | Coverage | Respond | Resolve or mitigate |
|---|---|---|---|---|
| P1 | Production down or data at risk | 24×7 on-call | < 15 min | Continuous until mitigated |
| P2 | Degraded or partial impact | 24×7 on-call | < 1 hour | Same business day |
| P3 | Bugs, requests, questions | Business hours, IST | Next business day | Scheduled |
Targets are defaults for a standard retainer; the exact tiers, hours, and escalation path are set per engagement and written into the agreement.