Louison Ribouchon
About
Hi! I’m Louison, an AI engineer living in the north of France. This is where I keep my projects, including the half finished ones and the ones nobody but me will ever open.
I came out of research. I spent a year in a lab in Kyoto building a vision model for a plastering robot, and I published from it, and somewhere in that year I realized I missed watching something actually get used. So I moved to product. I am now the only engineer on internal software and AI at an agricultural machinery company, which means I build whatever a department needs and then sit next to people while they try it, and I have ended up liking that second part quite a lot.
Most of my curiosity lately goes to agentic coding, and to the scaffolding around it that decides how carefully an agent works.
Outside of that I lift, I cook, and I collect Pokémon cards, and each of those has quietly turned into a project of its own.
Projects
Tracein progress
A workout tracker built around the set rather than the session, with an MCP server so an assistant can read your training history and talk to you about it. I open it every training day.
Neighbor-Aware ViT
A Vision Transformer that reads a wall tile by tile, looks at each tile’s neighbors before deciding, and tells a plastering robot where to work. It became my first paper, and I went to Xi’an to present it.
Synthetic defects for an optical sorter
An industrial sorter needed pictures of defects too rare to photograph, so I generated them with a diffusion model. We measured that against active learning and published both numbers.
reality-check
Open-source Claude Code skills for the moment you are moving faster with AI than you can check: audit the premises you were handed, declare where a fact came from, break out of a trial-and-error loop. Each one ships with its own evals.
philososkills
The same idea turned inward: six habits, each named after the philosopher who insisted on it. Verify before you assert. Try to refute your own answer. Look again at reality after a failure. They run on every task instead of waiting to be called.
YOLO for production lines
Vision models trained and deployed on agri-industrial production lines, and the dataset work underneath them, which is the part that decides whether a model still works a month later.
Contact
The fastest way to reach me is email. I read everything, and I answer.