Specialist, then Technical Specialist · Apple Retail
Parquesur, Madrid. Foundations in diagnosis, consultative support and daily execution under pressure.
jpp ~ ❯ fastfetch ┌────────────────────────────Hardware────────────────────────────┐ ▣ PC custom build │ ├▧ CPU AMD Ryzen 9 9900X3D (24) @ 5.58 GHz │ ├◆ GPU NVIDIA GeForce RTX 5090 │ ├▭ Display 3840x2160 @ 120 Hz in 42" │ ├▨ Disk 127.13 GiB / 1.82 TiB (7%) - btrfs │ ├▨ Disk 0 B / 1.82 TiB (0%) - btrfs │ ├◧ Memory 9.36 GiB / 60.46 GiB (15%) └ └⇅ Swap 0 B / 120.91 GiB (0%) └────────────────────────────────────────────────────────────────┘ ┌────────────────────────────Software────────────────────────────┐ ⌘ OS Omarchy 4.0.3-1 │ ├⌂ Channel stable │ ├⚙ Kernel Linux 7.2.3-arch1-3 │ ├⊞ WM Hyprland 0.56.2 (Wayland) │ ├◫ Terminal foot 1.28.0 │ ├▥ Packages 1045 (pacman) │ ├◐ Theme Tokyo Night └ └◨ Font JetBrainsMono Nerd Font (12pt) └────────────────────────────────────────────────────────────────┘ ┌─────────────────────Age / Uptime / Update──────────────────────┐ ◷ OS Age 36 years │ ├◔ Uptime 00:00:00 └ └↻ Update Sunday, September 13 2026 at 20:51 └────────────────────────────────────────────────────────────────┘ Type help to see the list of commands.
Meat-proxy · human-in-the-loop · slop-refiner
People-first technical problem solver.
Ten years fixing what nobody else could at Apple Retail. Same reflex now, aimed at agents, tools and small products that actually ship.
I sit between people and their machines: find what broke, say it straight, leave you with a next step that works.
01 / About
“The future is already here — it's just not evenly distributed.”— William Gibson
I am interested in the space between people and their machines: how agents and open-source tools can put real control back in the hands of the person running them — not the platform, and not whoever else thinks they have a claim on it.
A decade in the trenches taught me to diagnose fast, explain plainly and put the user first. Same goal now, different surface: agent harnesses, automation workflows and bespoke infrastructure that satisfy customer needs.
I want a machine that works the moment I turn it on, and doesn't fight me the moment I want to see how it works. Convenience and control aren't opposites — the best systems give you both: usable straight out of the box, still yours to open up when you need to. Open weights and open primitives matter for the same reason: whoever controls them decides where that ceiling sits.
For me, that means removing what is in somebody's way and leaving them with more control than they had before. It applies at a support bench, on a team and inside a technical system.
Metalhead, coffee problem, building and overclocking PCs since I was a teenager — the same instinct behind the homelab, GrapheneOS and getting off Google now. Dream car is still a Honda Civic.
02 / Experience
Apple Retail first, because it is the strongest evidence. Then the deliberate move into AI, and what I am studying while I make it.
I moved through Specialist, Technical Specialist, Genius and In-Store Experience Lead roles in Madrid. Most of the decade was spent as a Genius: advanced troubleshooting, repair triage, cases that had already defeated someone else, and explanations for people with very different levels of technical understanding.
I also worked as an AppleCare Support Advisor during COVID, handling remote cases under structured troubleshooting and documentation standards. As an ISE Lead, I ran store operations, resourcing and events, and dealt with the situations that did not fit a procedure.
The useful lesson was simple: the technically correct answer and the useful answer are not always the same sentence.
I moved from Madrid to Aberdeen and began a deliberate transition into AI and full stack development. I am learning by building: small applications, agent tooling, automation, deployments, and the documentation that explains what actually happened.
Parquesur, Madrid. Foundations in diagnosis, consultative support and daily execution under pressure.
Parquesur, Leganés. Six and a half years as the technical baseline of the store: advanced diagnosis on the cases nobody else had resolved, translating them for people with very different levels of understanding, and bringing newer technicians onto the bench.
Remote, during COVID, as a volunteer. Concurrent cases under structured troubleshooting and documentation standards, working without a floor around me.
Xanadú first, then Parquesur. Two seven-month leadership secondments taken while holding the Genius role: planning, resourcing, events, coordination and the situations that fit no procedure. Led by removing obstacles, then went back to the bench.
Moved to the UK and started a deliberate transition into AI and full stack development.
Studying at Universidad Rey Juan Carlos via Racks University, building Machines Do It Better as a solo practice with no clients yet, rebuilding a local-inference lab, and shipping small products that reach production.
Universidad Rey Juan Carlos, via Racks University (IUNIT-affiliated) · in progress
Universidad Rey Juan Carlos, via Racks University (IUNIT-affiliated) · in progress
Universidad Politécnica de Madrid · degree not completed
Apple
03 / Projects
Small products and agent tooling, shown with their limits and their receipts. Project notes say what I specified, wrote, deployed or built with an agent. The distinction matters to me.
Scheduled workflows, automations and agents with permission to touch real files. Skills are Markdown I can edit, tools are servers I can read.
My daily driver alongside Arch and Fedora. A desktop that expects to be edited rather than tolerated — the machine has to be shapeable before an agent can do much with it.
A private browser-based life-expectancy visualisation — useful as a prompt to act, without pretending to predict anyone in particular. No backend, no accounts, no analytics. Live on a VPS I administer.
MCP tools and agent skills for working safely with local Markdown vaults. Includes a tested filesystem layer and path-traversal protection.
An opinionated shortest path to a working Hermes Agent. It reduces the first hour of setup to one clear next step, instead of making a new user choose between every possible provider, model and configuration.
An unofficial landing page for Hermes Agent that behaves like the CLI it describes: a real command palette, six skins and no unnecessary build machinery.
Thirty machine-readable mental models for decision-making and risk checks, packaged as an agent skill pack and schema-validated in CI.
A nightly local paper for Aberdeen. Python computes the figures before the model writes the copy, so the model is never asked to invent the number. The architecture is upstream; my contribution is the city, the feeds and the editorial rules.
04 / Skills
Split three ways on purpose. Nothing in the second or third group is experience I have already had. The third is what I am starting next — learned with an LLM and an agent harness beside me, aimed at the 20% that makes a system operable: enough to deploy it, read what the harness proposes, and catch it when it is wrong.
05 / Achievements
Ten years in Apple Retail. Genius and ISE Lead experience. Small products and agent tooling built, documented and deployed while I learn.
Five roles across a decade, six and a half years of it as a Genius — the technical baseline of the store.
Two seven-month leadership secondments, both taken while holding the Genius role, then handed back.
A small app of my own, live on infrastructure I administer, behind an eight-check release gate that refuses to publish on red.
Agent skills, MCP tools, an onboarding guide and a news agent — each one stating plainly who wrote what.
Two programmes at Universidad Rey Juan Carlos via Racks University, both running to December 2026.
06 / Newsletter
Notes from the machine: agents, sovereignty, open systems, learning in public, and the strange work of making technology answer to its owner.
I prefer computers that can be edited rather than merely tolerated. Plain configuration, readable files and an owner who can understand the system are practical advantages, not aesthetic decoration.
I operate Hermes Agent daily and build tools and skills around it. The interesting question is not whether an agent can produce a convincing answer. It is whether the workflow is inspectable, the permissions are clear and failure stays visible.
Hermes Agent is built by Nous Research. My work here is usage, tooling, experimentation and documentation — not authorship of the underlying project.
I am rebuilding a two-rig local inference setup after a period offline. It is a real laboratory and a learning project, not a production endpoint anyone can call today.
The surrounding work covers Linux, Bash, networking, SSH, Docker, model serving, and the unglamorous parts of keeping a machine understandable.
07 / Contact
I am based in Aberdeen, Scotland. Remote work suits me. I am looking for work around technical support, customer-facing engineering, agent operations, implementation and practical AI tooling.
Independent, so no notice period. Aberdeen, Scotland, by way of Madrid, on UK hours. Remote suits me, I travel without fuss, and would relocate if it made sense for my partner too.