08:00 · Doors open

Welcome to the workshop.

I’m Raffael. I build open source, robots that find their own way, models that decide and interfaces people enjoy, and I am happiest learning something new. Spend a day in here with me and Abby, my French bulldog, who runs quality control.

01Tool wallOpen source02Robot arenaRobots + physical AI03Model labLLMs + decision models04Paint shopUI + landing pages05Software factoryFactory06Production logCV

The whole room is drawn in code.

  1. 01Tool wall09:30
  2. 02Robot arena11:00
  3. 03Model lab13:30
  4. 04Paint shop15:00
  5. 05Software factory16:30
  6. 06Production log17:30

09:30 · 01 · Open source

Useful tools, out in the open.

Everything I can share ends up on this wall: small libraries, CLIs and templates I needed myself first. Two outlines are empty. Someone forked them, which is the point.

Abby · QC Who is Abby?Abby, the real one: a brindle French bulldog in the grass, looking up into the camera.Abby · QC · the originalAbby is my French bulldog and my little helper. She keeps me busy, gets me away from the screen for our walks and has an opinion on every build. In here she is a robot, like everything else in the workshop.
Checked the wall. Two wrenches missing again. Apparently that is a good thing.

11:00 · 02 · Robots + physical AI

Machines that find their own way.

By day I lead frontend engineering at AGILOX, which builds autonomous forklifts that organise themselves as a swarm, with no central computer. My team builds what people see of them: the analytics for the whole fleet, live remote access to a vehicle, and a 3D replay of what it did and why. I also lead our AI agent, which lets operators ask the fleet questions in plain language.

02 · Field test

Next

Abby is aboard. Drive out through the door under the whiteboard.

Arrow keys or WASD: hold to speed up. Or tap the floor: near to crawl, far to go fast. Esc to leave.

Abby · QCThe robot reached the dock before me. I let it win.

13:30 · 03 · LLMs + decision models

Models that write, models that decide.

Language models are great at writing. For choosing, I like typed decision models like Jev: a state and a typed question go in, probabilities for each option come out, in one pass.

state + “what now?” → jev → one pass, ~0.2 s
ship0.86
wait0.10
ask0.04
Abby · QCAsked the model if it is snack time. 0.97. Nobody listened.

16:30 · 05 · Factory

A spec goes in. Reviewed, tested code comes out.

Downstairs is the factory. Briefs come in through the hatch, the press turns a typed spec into the predictable parts of an app, and build agents work the line while every change passes the review tunnel. What fails goes round again, and every project ends with a list of what went wrong that rides the green belt back into the template.

Abby · QCSits by the lead console. Nothing merges without a sniff.

17:30 · 06 · CV

Everything that left the workshop.

At the end of the day the label printer by the door prints the production log: one numbered entry for every role, launch and build. It is also my CV.

Production logRP · Linz
  1. Nº 001111.2026Freit: an AI transport office, a person makes every call
  2. Nº 001011.2026OpenFolia: manuals for every variant and language, open source
  3. Nº 000904.2026Team lead, frontend engineering · AGILOX
  4. Nº 000803.2026Released agentest: open-source tests for AI agents
  5. Nº 000710.2025AI engineer · AGILOX: fleet data, asked in plain language
  6. Nº 000604.2025Software architect, frontend · AGILOX
QC: Abbyapproved
Abby · QCStamped every entry. Quality control passed. Mostly.

17:45 · Shift notes

Written down before I forget.

The blog. Notes from the stations: what worked, what broke and what I would try next.

All notes

Commissions · a few a year

Sometimes I build for other people too: product prototypes, landing pages and help with AI and robotics projects.

Ask about a commission