Most organizations still work with competency models that are essentially built around terms. Job profiles, learning platforms, and HR systems use words such as “communication skills,” “leadership,” “problem-solving,” or “digital competence.” These terms matter, but they often remain vague. Two companies may use the same skill label while referring to completely different behaviors. Two managers may assess the same competency while applying different expectations, contexts, and personal interpretations. This is exactly where SupraTix’s innovation begins: competency is no longer treated merely as a text label, but as a scientifically validated, vectorized Behaviour Anchor within a Knowledge Graph.
A node in this model is not simply an entry in a skill database. It is a GraphRAG node that describes a concrete Behaviour Anchor — an observable behavioral pattern through which a competency becomes visible. This transforms an abstract term into a semantically precise, machine-readable, and comparable unit. The concept builds on an established tradition in work and organizational psychology. As early as the 1960s, Smith and Kendall developed behaviorally anchored rating scales, in which rating anchors were constructed around concrete examples of expected behavior to make assessments more specific and less ambiguous. Later research on Behaviorally Anchored Rating Scales also showed that such behavior-based anchors are particularly valuable wherever evaluation needs to be not only psychometrically sound, but also qualitatively meaningful and practically applicable.
The decisive difference is that SupraTix transfers this idea into the architecture of modern AI. Behaviour Anchors are not merely described; they are modeled as nodes in a Knowledge Graph ontology. Each node carries a semantic vector representation and can therefore be related to other nodes. Competency is no longer just a concept, but a calculable position in a semantic space. This logic aligns with what Knowledge Graph research has described for years: Knowledge Graphs are particularly suitable for structuring heterogeneous, dynamic, and relational bodies of knowledge through entities, relationships, ontologies, queries, and validation logic.
This is where the disruptive quality emerges. Traditional competency models are often static, organization-specific, and text-based. They are created in workshops, maintained in documents, and then transferred into HR systems. The result is often a catalogue, but not an intelligent infrastructure. A vectorized competency model fundamentally changes this logic. It makes it possible to computationally analyze similarities, distances, clusters, development paths, and competency gaps. Research on Knowledge Graph embeddings has shown that entities and relationships in multi-relational knowledge structures can be modeled in vector spaces. TransE, for example, represents relations as translations between embeddings and has been used for link prediction in large knowledge bases. Methods such as node2vec further demonstrate how nodes in networks can be translated into continuous representations through flexible random-walk strategies, making neighborhoods, roles, and structural similarities machine-usable.
For competency development, this marks a radical step forward. “Teamwork,” “conflict resolution,” “active listening,” “stakeholder communication,” or “fostering psychological safety” are no longer isolated items placed next to one another. They can be connected semantically, relationally, and developmentally. The system can identify which Behaviour Anchors are close to each other, which competencies are prerequisites for others, and which development steps make sense for an individual, a team, or an organization. Competency is no longer merely documented; it is operationalized.
This scientific foundation is crucial because competency models only become truly effective when they are properly analyzed, structured, validated, and made practically usable. Campion and colleagues describe competency modeling in Personnel Psychology as a field in which the analysis of competency information, its organization, and its practical use must be systematically connected. This is exactly what a Behaviour-Anchor-based GraphRAG approach enables: it combines organizational psychology with semantic AI, Knowledge Graphs, and generative language processing.
The innovation becomes especially powerful through its connection with Retrieval-Augmented Generation. Traditional Large Language Models can process language extremely well, but they are not automatically grounded in a validated knowledge structure. RAG research shows that language models can produce more specific, fact-oriented, and updatable responses when they have access to external knowledge stores. A GraphRAG model for competencies goes one step further: it does not simply retrieve text passages; it navigates through a curated competency network. The language model can therefore access validated Behaviour Anchors, semantic relationships, and explainable competency paths.
This makes AI in competency management more explainable. A talent-matching system does not merely claim that a person fits a role; it can show which Behaviour Anchors support that fit. A learning system does not simply recommend courses; it can derive learning paths from actual competency distances. An assessment system does not only produce scores; it can connect evaluations to observable behavior. A workforce planning system does not merely count roles; it can reveal which competency nodes are missing in teams, functions, or future scenarios.
At SupraTix, this approach did not emerge from a short-term AI trend. It is the result of eight years of research, development, and application experience. This long period of maturity matters because a globally applicable competency model cannot be created by simply merging skill lists. It requires scientific modeling, technical infrastructure, practical testing, validation, and the ability to describe human behavior in a way that is understandable to people and usable by AI systems.
A special milestone in this journey is the approval of the German Research Allowance and the award of the BSFZ Seal. SupraTix received the BSFZ Seal from the German Certification Body for Research Allowance, officially confirming that SupraTix conducts research and development within the meaning of the German Research Allowance framework and meets the requirements to claim the tax-based research incentive. This is more than an administrative step. The Research Allowance process requires an assessment of whether a project qualifies as eligible research and development. For SupraTix, this approval is therefore a recognition of its research work. It shows that the development is not only technologically ambitious, but has also been acknowledged as a genuine research and development achievement.
This is precisely where the disruptive force lies. SupraTix is not simply replacing an existing HR tool with a more modern interface. The real innovation goes deeper: the basic unit of competency management changes. The center is no longer the job title, the job description, or the simple skill list. Instead, the scientifically validated Behaviour Anchor becomes the smallest meaningful, observable, and AI-processable unit of competency. A taxonomy becomes an ontology. Text becomes vector. A static model becomes a dynamic graph. Subjective interpretation becomes explainable semantic relation.
This development is part of a broader research movement that sees Large Language Models and Knowledge Graphs not as opposing technologies, but as complementary ones. LLMs are strong in language, generalization, and interaction. Knowledge Graphs are strong in structure, explicit knowledge, relationships, and interpretability. Research on the integration of LLMs and Knowledge Graphs describes three central directions: Knowledge-Graph-enhanced LLMs, LLM-augmented Knowledge Graphs, and synergistic systems in which both technologies work together for knowledge-based reasoning. This is exactly where SupraTix’s GraphRAG competency space is positioned: as a bridge between validated knowledge, human behavior, and generative AI.
For organizations, this opens up a new level of strategic capability. Recruiting can become more precise because candidates are matched not only through keywords, but through behavior-based competency profiles. Learning can become more adaptive because recommendations are derived from actual competency distances. Leadership development can become more objective because feedback is anchored more strongly in validated Behaviour Anchors. Workforce planning can become more strategic because companies can understand not only which roles they have today, but which competencies they will need tomorrow. And AI systems in HR and learning can become more transparent because their recommendations are based on a validated semantic model.
The core innovation can therefore be stated clearly: SupraTix is developing a scientifically validated, vector-based competency model in which each GraphRAG node represents a Behaviour Anchor. This creates a globally applicable competency space that does not merely name human capabilities, but makes them observable, comparable, calculable, and operationalizable for AI. The approval of the Research Allowance and the BSFZ Seal underline that this approach is built on real research and development work. For us, this is a recognition of our work because it shows that this innovation is not just another AI use case. It is a new infrastructure for measuring, developing, and connecting human competencies in the age of generative AI.
