Linux and production operations
Debian/Ubuntu and RHEL-family: systemd, hardening, patch policy, and diagnosis with ss/ip/nstat rather than guesswork.
20+ years in production infrastructure: hypervisor and container estates, the network paths between them, and the tooling that makes changing them safe.
Cork, Ireland · EU citizen · one month's notice
Six things, each with the ceiling that comes with it.
Debian/Ubuntu and RHEL-family: systemd, hardening, patch policy, and diagnosis with ss/ip/nstat rather than guesswork.
Virtuozzo/OpenVZ container hosting and VMware vSphere in production; KVM underneath both, Proxmox VE on my own estate.
Ansible with an executable inventory and reviewed changes, plus CI that fails the build instead of printing a warning nobody reads.
Routing, DNS, VPN, NAT, firewall zones, reverse proxy and TLS termination.
Incident ownership, blameless root-cause analysis, runbooks, and restore rehearsal treated as the real backup control.
AWS Solutions Architect - Professional. Azure exposure is limited to roughly 2-3 months, and I say so rather than implying otherwise.
AWS Certified Solutions Architect - Professional · issued Oct 2024, valid to Oct 2027
The two roles that describe the last three years.
Production cloud and hybrid infrastructure across two data centres: Linux hosts, Virtuozzo/OpenVZ containers and the interconnect between them. Incident owner and escalation point on a 24/7 shift-covered service. Previously Cloud Lead Engineer / Architect on the same platform, employed in Cyprus.
SME and private engagements: DNS and domain architecture, VPS and Linux hosting with security applied at install time, CDN, reverse proxy and TLS termination.
Self-directed engineering, graded by evidence rather than by adjective.
Five roles, seven playbooks and an executable inventory, with validation that fails the build rather than warning: lint escalation, inventory parsing, two-tier secret scanning, and a regression guard for the lint config itself.
Verified | code and local run View evidenceVLAN-aware bridges, isolated guest bridges, SDN/VNet isolation and firewall-path audits on a self-directed estate, with OpenTofu provisioning that validates cleanly, and has not been applied.
Lab View evidenceTunnel-based remote access with internal services kept private, expected-state assertions with a drift check, and a DNS chain of forwarder → DNSSEC validator → encrypted upstream. The OpenWrt control plane itself is designed, not cut over.
Verified | code · OpenWrt proposed View evidenceOne authenticated endpoint over 53 tools (GPU inference, RAG memory, Docker, browser), with every backing service kept LAN-only, per-principal tool ACLs, and published latency and guardrail measurements.
Verified | code (private) View evidenceDependency-free Python parsing research: deterministic extraction with typed failure codes, a fact ledger with two hashes per fact, and a source register that grades every claim A/B/C/U. No score is offered, because the A-graded evidence it cites shows a ranking function of this class reverses 29.6% of competence-preserving pairs.
Verified | code View evidenceThe path this site is built by: a private source repository, a checker that proves it can fail before it is allowed to pass, a 15 ms Hugo build, and a one-way force-publish to Pages. 62 source files in, 11 artifact files out, 6 green runs.
Verified | code (private) View evidenceDocument, code, lab or proposed. Nothing is published above what its artifact supports. The audit is public, including what is deliberately not claimed.
I adopt a technology by running it until I can name its failure modes, then write down which ones I have actually met. Where something is planned rather than deployed, it says so on the page that describes it.
Every claim here carries an evidence tier (document, code, lab, or proposed). Nothing is published above what its artifact supports. How the claims are graded →