CI/CD · SRE · RELEASE ENGINEERING
I run build, deployment and test environments for R&D from within the QA team, plus monitoring, reverse proxies and internal systems. Since mid-2025 I develop with AI agents and control their output with automated tests and hands-on verification. Every case states my scope, how it was verified, and where each number comes from.
Delivery
Run
INFRASTRUCTUREProxmox VE · TrueNAS · OPNsense · Linux ×100+
ROLE & SCOPE
The delivery and operations engineer inside QA: R&D writes the code; I own the pipeline and infrastructure from merge to test, release and monitoring.
| Product / system | My responsibility | Period | Related hours | Evidence |
|---|---|---|---|---|
| Company cloud productR&D test environment | GitOps deployment via GitLab CI → Harbor → ArgoCD; RKE2 cluster, Envoy Gateway routing and TLS | 2024-10 – now | 362 h (me ≈175 h · AI ≈187 h) | Deploy repo: 419 commits (117 mine) · Read case |
| R&D build environment6 build lines incl. Web, Go, Java, Windows | GitLab Runners and build hosts; moved Jenkins into containers, later migrated its pipelines to GitLab CI/CD | 2021-04 – now | GitLab CI 472 h (me ≈427 h · AI ≈45 h) · Jenkins 681 h | 6 CI/CD handover manuals (Sep 2026) |
| Public-facing reverse proxyPublic company websites | Rebuilt the proxy and WAF, config under GitOps, scheduled upkeep with Ansible | 2022-10 – now | 939 h (me ≈843 h · AI ≈96 h) | 1,425 of my commits to the config repo · Read case |
| Monitoring platformCompany hosts and the K8s cluster | Grafana, Prometheus, Loki, Alertmanager; per-team scoped views | 2022-06 – now | 496 h (me ≈422 h · AI ≈74 h) | 58 alerting rules, 22 recording rules, 18 scrape targets · Read case |
| Internal systemsProject management, image registry, collaboration tools | Upgrades, data migration, incident investigation and reports | 2022 – now | Project management 126 h (me ≈76 h · AI ≈50 h) | Incident reports and operations log |
| Identity verification serviceCloud gateway | Platform side: namespaces and RBAC, gateway routes, restricted deploy identity; the service team owns Helm values | 2025-12 – now | — | Self-service guide · handover manual |
Every result states its source. Where nothing was recorded, no number is given.
CASES
Each covers context, my scope, verification, results and limits
The product runs as containers on an RKE2 cluster used as R&D's test environment. After a merge, GitLab CI builds images and updates a deploy repo; ArgoCD syncs the cluster. Nobody logs in to a host.
Rebuilt the public reverse proxy, hardened it with ModSecurity (OWASP CRS) and Fail2Ban, put every config change under Git, and surfaced block events in Grafana.
Built the company's monitoring from scratch: Grafana, OpenSearch and Fluent Bit first, then Prometheus, Loki and Alertmanager, with views scoped per service team and alerts by email.
Turned existing manual test cases into Gherkin, Robot Framework and Playwright scripts. Every step keeps before/during/after screenshots and a recording; the product QA reviewed each item before taking over maintenance.
Building Kubernetes clusters from scratch and repeated lab work exposed their complexity and learning curve. To help IT staff provision and configure clusters, I began developing a Go management platform with AI agents in August 2025. A VM-based on-premises cluster on Proxmox takes an estimated 20–30 minutes, depending on storage I/O; backup and recovery capabilities are verified separately.
AI-ASSISTED ENGINEERING
I first paid US$20 a month for AI tools myself; when Claude Code arrived I moved to US$100 a month from 2025-07-09 and began developing with AI agents. Development work, commits and new projects all rose sharply afterwards. My own hands-on share of the time fell from about 80% to about 20% (own estimate).
MeAI Agent
Early on, steady output from agents took a lot of hand-written scaffolding: CLAUDE.md/AGENTS.md, rules under .claude, SKILL.md files and project indexes, plus my own loop and harness engineering. With each new Claude generation more of that became built in. My time moved to decisions on data control and flow, system architecture and maintainability, while agents handle implementation and execution.
Ratios are my estimate, stepped at Claude model releases. Each case’s logged hours are split by the period’s ratio (at 2:8, 100 h is 20 h mine and 80 h AI); hours before 2025-02-22 count as mine. From 2026 models also run overnight, outside the timesheet, so only company-logged hours are counted and actual AI time is higher.
I define requirements, split tasks, make the data-control and architecture decisions, review and accept; agents draft code, tests and docs.
AI output is controlled by tests, screenshots, recordings and hands-on checks; all code and config goes through version control and review.
TIMELINE
The last entry each year groups short projects, ad-hoc testing and POCs not listed separately; hours come from timesheets and exclude leave, admin and meetings.
SCOPE
| Layer | Tools | What I did |
|---|---|---|
| Delivery |
| Runners and build hosts, GitOps deploys, Harbor upgrades, Jenkins migration |
| Containers |
| RKE2 cluster build; K8s 1.29 → 1.34 upgrades; Ingress Nginx to Gateway API |
| Observability |
| Alert rules and notices, per-team dashboards, central logs |
| Virtualisation and network |
| Disk I/O and storage incidents, recovery after outages, security events |
| Automation and testing |
| Ansible operations automation; product test suites; Go projects written by AI agents, reviewed and verified by me |
EXPERIENCE
R&D build and deploy pipelines, the K8s test environment, monitoring, the public reverse proxy and internal systems; product test automation, release testing and second-line verification.
Ran customer systems and websites; introduced vulnerability scanning and patched against advisories; automated operations with Bash and PowerShell.
IBM Power server and storage replacement, AIX maintenance and upgrades, restore drills with customers.
Deployed and maintained security products, wrote product documentation, supported customer POCs.