ESCOX: the news kills extractor tool
In today’s fast-changing job market, staying aligned with the right skills is more important than ever. Yet identifying what skills are in demand, and how they connect to actual jobs, is still a challenge. Traditional keyword searches often fail to capture the complexity of job descriptions or match them to recognised taxonomies like ESCO (European Skills, Competences, Qualifications and Occupations).
That’s why the SKILLAB Horizon Europe project developed ESCOX: a new open-source tool designed to extract both skills and occupations from unstructured text using Large Language Models (LLMs). The tool helps educators, policymakers, and HR professionals better understand the skills landscape and make more informed decisions.
What is ESCOX?
ESCOX (ESCO Skill Extractor) is an AI-powered tool that reads job postings, policy documents, or reports and automatically identifies and classifies the relevant skills and occupations mentioned in the text. Unlike many existing tools, which focus solely on skills, ESCOX recognises both what individuals can do and the roles they are qualified for, providing structured outputs aligned with both ESCO and ISCO-08 standards.

How does it work?
ESCOX works by reading unstructured text and automatically identifying the skills and occupations mentioned. It does this by comparing the words in the text to a large database of official skill and occupation categories, using artificial intelligence to understand the meaning rather than just looking for keywords. Users simply paste a piece of text into the tool or upload a file, and ESCOX returns a structured list of relevant skills and job roles, complete with confidence scores. The tool can be used through a simple web interface, requiring no technical background, or integrated into more complex workflows via its open-source software and API. This makes it accessible to both non-technical users and developers who want to analyse large volumes of data.
Why it matters
Using real-world job data from the European Employment Services, ESCOX was tested on over 6,500 job postings. The tool identified high-demand technical skills like Java and SQL, but also key soft skills such as communication and problem-solving. In parallel, it recognised roles like ICT Business Analyst and Project Manager, making it easier to map demand in the digital sector.
This kind of insight is valuable for:
- Policy design: identifying regional or sectoral skill gaps
- Curriculum development: aligning educational content with real-world demand
- Recruitment and workforce planning: better job descriptions and talent strategies
ESCOX is highly relevant for initiatives like TechConnect, which focus on the interaction between human capabilities and emerging technologies. One of TechConnect’s core goals is to understand how digital tools can support skills development, workforce planning, and more inclusive labour market strategies. ESCOX contributes directly to this by making it easier to map skills and occupations from real-world data in a structured and scalable way. It helps bridge the “affordance gap” between what technologies are designed to do and how they are actually used, by providing clear insights into the evolving language of work. As a tool that supports data-driven decision-making around education, employment, and policy, ESCOX is well aligned with TechConnect’s mission to build human-centred technology ecosystems.
Built for usability and scale
ESCOX has been designed to be both accessible and technically robust. It includes a user-friendly web interface that allows non-technical users, such as policymakers or HR professionals, to interact with the tool easily. The system can also be deployed using standard infrastructure and performs efficiently even without GPU acceleration. It supports multilingual input, integrates summarisation mechanisms for processing long documents, and can be accessed via API or run in environments such as Google Colab.

Looking ahead
Future versions of ESCOX will include deeper learning models, improved processing speed (e.g. using FAISS), and new visual features to make the results even easier to understand. These upgrades will help further improve the quality of decision-making in education, training, and employment policy.
Read the full paper here: https://www.sciencedirect.com/science/article/pii/S2665963825000326?via%3Dihub