AI engineer. Database and infrastructure specialist.
Senior Oracle DBA at Atos. Developer, DBA, team leader and manager across finance, publishing, telecoms, utilities and the public sector. I also build AI apps and agents, and the servers they run on.
United KingdomDownload as PDF
The day job. Oracle 19c with Data Guard, Transparent Data Encryption, RMAN and Enterprise Manager on Red Hat in Azure. Bash automation for application patching, and a daemon that runs a client's batch processing. Low-level designs, standards and procedures for each build.
Building automazr taught me the modern stack end to end. I moved a Next.js app off Vercel and Supabase onto a server I provision entirely from code: Fedora CoreOS with Butane and Ignition, rootless Podman quadlets, Postgres 18, and Caddy behind Cloudflare's proxy, WAF, Access and authenticated origin pulls. A separate host runs Prometheus, Loki, Alertmanager and Grafana, and alerts by email.
AI agents are part of how I work now. I use Claude Code, opencode, Pi and oh-my-pi from written specs, then review what they produce. They run sandboxed: one Nix-built container image holds all four, with capabilities dropped and only the project mounted, or a micro-VM. I run local models through Ollama and llama.cpp to learn where open models hold up at tool calling and where they don't.
Live. Sign-in required; ask me on LinkedIn for a demo.
An AI-assisted web app on infrastructure I built and run myself. Next.js, Postgres 18 with pgvector, Better Auth with enforced TOTP, and Gemini for suggestions, with a fallback for when the model isn't available. Written with AI coding agents from my specs, then deployed to the self-hosted, monitored host described above.
Demo available on request.
Fixed workflows or an AI agent: which runs a Kubernetes cluster better? tkube puts n8n and kagent on the same kind cluster in rootless Podman so you can watch both handle the same problems. The agent's model is an open model running locally on my AMD GPU through the official llama.cpp ROCm image, so nothing leaves the machine. The whole lab comes up and down with one command from a devenv shell, and I demo it live over a Cloudflare Tunnel.
In daily use on my desktop and laptop.
My own Linux desktop, shipped as a container image. It's built on Bazzite (Fedora Atomic) with the Niri compositor and Noctalia shell, and has AMD and NVIDIA variants. GitHub Actions builds both, the image is rechunked, signed with cosign and published to GHCR, and machines update by pulling the new image, with a one-step rollback if it misbehaves. It also carries Kata Containers, YubiKey tooling and Nix.