I work on AI products, engineering tools, and the systems around them.
I’m Arthur. I lead a small team of AI engineers at DxO Labs.
Our work spans several areas, but I would describe it simply: we rethink products, processes, and parts of the organisation so that people and AI can work through the same systems. We build internal tools, evaluation loops, engineering processes, and new features for photographers.
One strand of that work is a visual agent for creative software. I use vibe retouching as shorthand for expressing a visual intent in ordinary language while an agent translates it into a bounded, inspectable sequence of actions. The aim is to shorten the path to a first editable result without replacing craft or direct control.
Outside DxO, I co-founded DailySurf, where I work as CTO and lead product engineering. DailySurf is a mobile app for surfers to record their sessions and follow their progress over time.
Previously: mathematics and machine learning at ENS Paris-Saclay, followed by remote-sensing work at Kayrros, CNRS, and Inria.
Professional photo-editing software has a steep learning curve. We are exploring a visual agent that translates an ordinary-language intent into a controlled sequence of actions inside the software, while keeping the result editable by the user. I will write more about this soon.
Inspired by how teams at OpenAI and Anthropic work with coding agents, I set up a Linear-based workflow in my team. It connects human decisions, bounded agent work, checks, CI, and review; I now advise other R&D teams as they adapt it.
Outside DxO, I co-founded DailySurf and serve as CTO. I lead product engineering and have built most of the app, which helps surfers record sessions, remember conditions, and follow their progress.
A conversation with I-Academy about introducing AI into day-to-day work across a company. In French.
My speech at the 2023 ENS Paris-Saclay graduation ceremony begins at 1:44:00. In French.
Smartphones, AI, and the fact that products evolve faster than people do.
A personal account of leaving academia and choosing to work on products.
Hand-built seasonal features compared with satellite foundation embeddings.
Training land-cover maps to remain useful across dates, seasons, and biomes.