The graph in numbers.
Snapshot of May 14, 2026 · Updated monthly
The eight frameworks.
Why join what has never been joined before.
ESCO
The official European standard. Maintained by the European Commission. The preferred framework for mobility within the EU.
ROME
The France Travail framework (formerly Pôle Emploi). 532 job profiles, a backbone of French HR.
O*NET
US Department of Labor. The framework richest in behavioral data. Used by every US ATS.
COMPETENT
A French domestic, sector-based framework. Covers cross-functional professional skills that ROME does not.
Singapore SSF
Singapore's government Skills Framework. The reference in Southeast Asia, a model for SkillsFuture.
CO-SISPE
Spanish occupation framework (CO-SISPE). 2,319 occupations imported; skill extraction and cross-standard junctions are being integrated.
AI4ESCO
Italian occupation framework (AI4ESCO). 829 occupations imported; skills and cross-standard junctions are being integrated.
PT
Portuguese occupation framework. 664 occupations imported; skills and cross-standard junctions are being integrated.
Skill Trees.
A skill never stands alone. It has prerequisites.
Most skills graphs are flat lists: a bag of words with synonyms. The Skillberg graph adds a fundamental dimension — prerequisite order.
To master Deep Learning, you first need Matrix calculus, Python programming and Bayesian statistics. These prerequisites are not hard-coded: our algorithm generates them automatically from a corpus of training programs, job postings and expert validations.
This IP — Skill Trees and the algorithms that generate them — is what lets Skillberg answer questions no other platform can handle: “What is the shortest path for a PHP developer to become a Data Engineer?” or “Which training should I fund to close this skills gap in 18 months?”
Invented by Martin Vielvoye during his PhD in AI and neuroscience. At the heart of the Skillberg graph since day one.
How we built the graph.
Twenty-four months of doctoral R&D (CIFRE), three sources, two levels of validation.
Data creation
Building the Skill Trees: each skill is modeled with its prerequisites (an ordering graph). The proprietary structure that makes the graph queryable.
Raw ingestion
Import of the 8 official frameworks through their public APIs. Normalization into canonical URIs skillberg://skill/{uuid} and skillberg://occupation/{uuid}.
Cross-standard joining
A proprietary semantic matching algorithm across frameworks, combining embeddings + linguistic rules + official metadata. 18,073 junctions produced this way.
Expert validation
Every junction the algorithm proposes is reviewed by a pair: a domain expert and an engineer. Current validation rate: 87% of algorithmic proposals.
Full pipeline documented in the whitepaper. Data available for partial reproducibility on academic request.
Graph schema.
Three node types, six edge types. Stable since v1.0.
(:Skill {uri, label, referentials[], proficiency_levels[]})(:Occupation {uri, label, isco_codes[], rome_code, esco_code})(:Referential {name, version, last_sync})(:Skill)-[:PREREQUISITE_OF]->(:Skill)(:Skill)-[:BROADER_THAN]->(:Skill)(:Skill)-[:NARROWER_THAN]->(:Skill)(:Skill)-[:CO_OCCURS_WITH]->(:Skill)(:Skill)-[:JOINS_VIA]->(:Skill) // cross-standard(:Occupation)-[:REQUIRES]->(:Skill)
Sample query.
All direct and indirect prerequisites of Deep Learning:
Cypher access available on the Omni tierMATCH p = (s:Skill)-[:PREREQUISITE_OF*1..3]->(t:Skill {label:'Deep Learning'})RETURN s.label, length(p) as depthORDER BY depth ASC
Whitepaper: Building a canonical skills graph.
Architecture, methodology, validation metrics, reproducibility. 24 pages, in French.
Download the whitepaper (PDF)Explore the graph live.
The Skillberg console lets you navigate the graph interactively. Search, profiles, neighborhood, prerequisites. Free once you create an account.
Open the console