Why we exist.
The AI Act classifies AI systems used for hiring, assignment and evaluation as high-risk (Annex III, deadline pushed back to December 2, 2027): they will have to be explainable and overseen by a human. It is a major regulatory shift — and yet the vast majority of today's HR platforms still operate as black boxes.
We built Skillberg to provide the alternative. A canonical graph, joined across eight global frameworks, where every score, every match, every recommendation can be traced back to its source node. Not an explainability patch. A native property.
We built it in Europe, on European infrastructure (Scaleway), with European models (Mistral). Because HR data sovereignty is not optional: it is a contractual obligation to every employee whose data passes through our system.
“No critical HR decision should rest on a black box.”
— Martin Vielvoye, founder
The invention that started it all.
Before Skillberg, Martin Vielvoye was doing research in artificial intelligence and computational neuroscience. His subject: how to formally represent the structure of human knowledge — and in particular, how to model the fact that one skill unlocks others.
That work produced an invention: Skill Trees. Skill graphs directed by prerequisites, with self-generating algorithms that can compute, for any target skill, the shortest path from a learner's current state.
The core insight: no skill stands alone. Deep Learning means nothing without Linear algebra, Bayesian statistics, Python programming. Modeling them as a bag of words is an epistemological mistake. Modeling them as a prerequisite tree is what makes the strategic HR questions answerable — which training to fund, which mobility paths to open, which skill gaps are deal-breakers.
Skillberg is the direct commercialization of this invention. Our proprietary graph — 158,000 skills across 8 frameworks, 18,073 cross-standard junctions — is entirely structured by Skill Trees. It's our moat. It's what no competitor will replicate in less than two years.
CIFRE (industrial PhD) thesis in progress · Ongoing expert validation · KG whitepaper available at /ressources.
Seven people. One lab.
All based in France. No subcontracting. No outsourced juniors on engagements. The code, the graph and the research stay in-house.
A growing team. Explore our resources →
Milestones.
A few key moments.
Skill Trees invented
Martin Vielvoye's thesis formalizes the prerequisite model.
Skillberg founded
Company incorporated in Lille, hosted by Euratechnologies.
First canonical graph
ESCO + ROME + O*NET ingested, first cross-standard junctions.
BPI Innovation
BPI support for the R&D phase + hiring.
Public API launch
api.skillberg.app, OpenAPI 3.1, Builder tier.
4th organization at TalentCLEF 2026
Task A, résumé–job matching: organization ranking in the international CLEF benchmark.
MCP connector for Claude and ChatGPT
The Skillberg graph becomes available from Claude, ChatGPT and any MCP client.
Backers and partnerships.
Institutions, programs, infrastructure.