Organisations increasingly steer learning, staffing and workforce planning through skills frameworks: catalogues that define which skills exist, at which proficiency levels, and how they relate to job profiles and learning content. Skills-based talent management only pays off if the whole organisation works with one consistent skills language (Jooss et al., 2024). In practice, that language loses consistency over time. 

Scheer IMC, the industry partner of this project, provides the IMC Learning Suite, a learning management system in which client companies maintain their own skills catalogues linked to job profiles and learning content. A catalogue may already exist and need to be adapted and extended, or it may need to be modelled and constructed from organisational evidence. In both cases, maintaining an adaptable, “living” corporate skills framework is crucial. 

As roles change and new technologies emerge, internal company-specific taxonomies need to evolve, while external, more standardised taxonomies such as ESCO (le Vrang et al., 2014) are continuously updated. Otherwise, job advertisements, CVs, project documentation and course descriptions drift apart terminologically. Duplicates, near-synonyms and outdated entries accumulate and undermine skills matching, learning recommendations and workforce analytics. 

Automated skill extraction from unstructured text has matured considerably (Senger et al., 2024), yet a single extraction model does not solve the problem. Constructing and keeping a catalogue alive requires extraction, deduplication, normalisation, harmonisation of taxonomies and proficiency frameworks, job-profile generation and updating, as well as document tagging. Structural changes also carry governance consequences. 

The setting is therefore well suited to LLM-based multi-agent systems, in which specialised agents are orchestrated and critical decisions are routed to human reviewers (Guo et al., 2024; Tsaneva et al., 2025). The project asks how a multi-agent system can construct a corporate skills framework from organisational evidence and available reference frameworks and maintain it accurately and traceably as those sources evolve.

Project Objectives: The project develops a functional multi-agent prototype that can construct a corporate skills framework and continuously maintain it as organisational evidence and reference frameworks evolve. Starting from organisational source documents and, where available, an existing seed catalogue of skills, proficiency levels and job profiles, specialised agents construct and subsequently maintain the framework by analysing and monitoring internal sources and external reference frameworks. They propose additions, modifications, merges and deletions and route significant structural changes for human review. All applied changes remain traceable through the system’s change history. The target architecture comprises:

• Extraction agent: detects skills and indications of proficiency in unstructured text such as job advertisements, CVs and project documentation.
• Deduplication and normalisation agent: clusters similar or duplicate skills and proposes merges.
• Mapping agent: harmonises skills and proficiency information across internal vocabularies and external taxonomies such as ESCO or O*NET.
• Profile and tagging agents: generate and update job profiles and tag documents and learning content with the normalised skills vocabulary for semantic search.
• Orchestrator and review dashboard: coordinates the agents, resolves conflicting agent results, manages human-review steps, and visualises the living framework, its change history and open approvals.

Results are evaluated using measurable criteria, for example precision and recall of skill extraction, agreement of proposed taxonomy and proficiency mappings with expert-reviewed reference mappings, alignment of generated job profiles with expert-reviewed reference profiles, or catalogue consistency before and after normalisation. The choice of frameworks and programming languages is left to the group; Python in combination with an agent framework (e.g. LangGraph, AutoGen or a comparable framework) is recommended. 

Project Requirements:

• Creation of a project plan and allocation of tasks among the group members using project management techniques (standard/mandatory).
• Analysis of organisational data sources, relevant HR and learning processes, and interactions between the IMC Learning Suite and surrounding systems; requirements elicitation with Scheer IMC and derivation of a prioritised product backlog.
• Acquisition, construction and/or synthetic generation of the working data by the group: a seed skills catalogue and a realistic corpus of documents (job advertisements, CVs, course descriptions), together with relevant internal skills vocabularies, taxonomies and proficiency frameworks, plus acquisition of relevant external reference frameworks.
• Development of a working multi-agent prototype with at least three specialised agents and an orchestration layer.
• Implementation of a human-in-the-loop review step for significant structural changes, including change history and traceability of every applied change.
• Measurable evaluation of the prototype, e.g. precision and recall of skill extraction, agreement of proposed taxonomy and proficiency mappings and generated job profiles with expert-reviewed references, or catalogue consistency before and after normalisation.
• Project documentation and presentation of the results in interim and final presentations (standard/mandatory).

Prerequisites:

• Students of the Faculty of Computer Science in the fields of Information Systems or Software Engineering at bachelor's level (with at least 90 ECTS) or at master's level (standard/mandatory)
• Solid programming skills (preferably Python) and willingness to work with LLM APIs and agent frameworks
• Interest in HR and corporate learning processes and in regular coordination with the industry partner

Literatur

• Guo, T., Chen, X., Wang, Y., Chang, R., Pei, S., Chawla, N. V., Wiest, O., & Zhang, X. (2024). Large language model based multi-agents: A survey of progress and challenges. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI-24) (pp. 8048–8057). doi.org/10.24963/ijcai.2024/890 
• Jooss, S., Collings, D. G., McMackin, J., & Dickmann, M. (2024). A skills-matching perspective on talent management: Developing strategic agility. Human Resource Management, 63(1), 141–157. doi.org/10.1002/hrm.22192 
• le Vrang, M., Papantoniou, A., Pauwels, E., Fannes, P., Vandensteen, D., & De Smedt, J. (2014). ESCO: Boosting job matching in Europe with semantic interoperability. Computer, 47(10), 57–64. doi.org/10.1109/MC.2014.283 
• Senger, E., Zhang, M., van der Goot, R., & Plank, B. (2024). Deep learning-based computational job market analysis: A survey on skill extraction and classification from job postings. In Proceedings of the First Workshop on Natural Language Processing for Human Resources (NLP4HR 2024) (pp. 1–15). https://doi.org/10.18653/v1/2024.nlp4hr-1.1 
• Tsaneva, S., Dessì, D., Osborne, F., & Sabou, M. (2025). Knowledge graph validation by integrating LLMs and human-in-the-loop. Information Processing & Management, 62(5), 104145. doi.org/10.1016/j.ipm.2025.104145 

If you are interested in this porject, follow the steps below to submit your application.

1. Form a project group
Projects are typically carried out in groups of 3–5 students. We recommend forming a group with fellow students before applying.

2. Prepare your application
Send a short application including:

• your transcript of records
• a short motivation letter (about one page) explaining why your group is interested in the project.

3. Submit your application
Send your application via email to:mailto:jannis.nacke (at) icb.uni-due.de