The lab
About the lab
Who
AME Labs is the engineering practice of Vincent Grandjean, an independent researcher in Paris. The road here ran through nearly three decades as an operator — industry, consulting, founding and running companies across Europe and Africa. The interest underneath is older: a TRS-80 Color Computer at twelve, and the shelf of science fiction — Asimov, Clarke, Banks — that made artificial minds feel like a future worth building. After the business career, this lab is the return to that first love. Since 2025 he has been building AME-1, a specification-first framework for long-lived, local, governed personal AI characters — and measuring it in public.
The lab is currently one researcher, working with AI tooling — which is part of the research thesis, not a disclaimer. How far a disciplined solo practice can carry a full cognitive architecture, with modern AI as the workforce, is itself one of the experiments.
How we work
The lab’s claims are designed to be checkable rather than impressive. The method, in order:
- Literature first. Grounded in the AI and neuroscience literature — the lab researches existing science, then engineers from it.
- Specification second. The architecture is written down before it is built — data models, algorithms, typed event catalogs for every module, process specifications, test plans, and dated architectural decision records — machine-checked for coherence after every change.
- Code third. The build follows the specifications module by module, against hard gates — the specifications’ test plans checked by tooling.
- Strict experimental evaluation fourth. Pass/fail bands are frozen before the test questions exist, both outcomes pre-named; workloads and knowledge graphs are pinned by git hash, and a post-freeze edit voids the measurement; expected outcomes are derived from the shipped code and predicted in writing before every run.
- Publish the map. The architecture and the reasons are public; the claims stop where the measurements stop.
- A modern stack. Rust, PostgreSQL, a three-tier event-driven architecture, multi-model local inference — local models via Ollama, in-process ONNX.
What “lab” means here
An independent engineering practice — not, yet, an incorporated entity. The purpose is a product, and the register is evidence-first: published articles, measured results, a lab notebook. Research conversations, sponsored research, and grant discussions are welcome: contact.
Why Europe
The lab’s philosophy and Europe’s are the same bet. Privacy as a structural property, not a promise; data that stays on hardware you own; systems accountable for what they do, and able to show why. These are the principles this lab engineers from — and they are the principles Europe has chosen to stand for, in law and in ambition: sovereignty over one’s own data, AI that answers to the people it serves. AME Labs builds in Europe, under European law, because here that philosophy is native.
The mark — the thinker, or the soul in the machine
A square labyrinth — the machine — whose last corridor rises and resolves into a figure bowed in thought: the human in the machine. Every line in the drawing is rectilinear except one; the red path is the only living thing in the structure, and it is the same red that marks the answer rail in the lab’s figures. In French, âme is the soul — AME-1 reads as âme un. The thesis, in one drawing.