Data Maintenance Made Easy - AI to Support Our Personnel Profile Platform

As a consulting firm for complex IT projects, we don’t just develop customized solutions for clients - we also drive our own product ideas forward. One of these developments is Profilery, a platform for the digital management of employee profiles - designed for organizations that regularly need to create and maintain personnel profiles for client projects.

To simplify the manual maintenance of these profiles and simultaneously improve data quality, we strategically leverage Artificial Intelligence. Together with the Mittelstand-Digital Zentrum Augsburg, we questioned existing AI ideas, developed new approaches, and deepened our expertise.

The Mittelstand-Digital Zentrum Augsburg supports small and medium-sized enterprises in harnessing the opportunities of digitalization. With practical offerings, the center imparts knowledge about digital technologies - from Artificial Intelligence to Industry 4.0. The focus is on tangible added value and implementability in everyday business.

#Our Platform: Centrally Managing Digital Competency Profiles

Our SaaS solution Profilery supports companies - such as staffing agencies or consulting firms - in capturing, maintaining, and managing the qualifications and experiences of their employees. The profiles can be maintained centrally on a digital platform, which accelerates the entire proposal process and makes it more transparent.

A key goal: sales staff should be able to access a reliable, up-to-date data foundation at any time to select suitable personnel for client projects.

#AI in Action: Automation and Quality Assurance

AI is already deployed at several points within Profilery:

• Simplified data entry, e.g., in type-ahead search and in generating individualized suggestions through LLMs
• Intelligent filter and search functions for targeted candidate selection
• Automatic categorization of competencies into thematic clusters

These features make it easy to maintain competencies, find specific skills, and systematically structure data. But we want to go further. That’s why new approaches were developed in dialog with the Mittelstand-Digital Zentrum Augsburg. The AI experts Dr. Martin Gottwald and Alexandros Tsakpinis from fortiss examined the platform closely, reviewed the existing ideas in detail, and critically challenged them as sparring partners with an “outside perspective.”

fortiss is the Bavarian State Research Institute for software-intensive systems. Supported by the Free State of Bavaria and the Fraunhofer-Gesellschaft, fortiss operates as a scientific affiliated institute of the Technical University of Munich. The institute conducts application-oriented cutting-edge research in the areas of Software & Systems Engineering, AI Engineering, and IoT Engineering. It serves as a bridge between science and industry and develops solutions for challenges in various sectors such as automotive, energy, aerospace, manufacturing, finance, insurance, public administration, and healthcare.

#Assistance System for Semi-Automated Profile Creation

The manual maintenance of competency profiles often falls short in everyday work. Therefore, we are pursuing the idea of an AI-powered assistant that supports users in completing their profiles - comparable to a digital advisor for improving consistency, completeness, and data quality.

Such an assistance system can:

• Offer suggestions for similar or frequently used competencies,
• Point out potentially “forgotten” competencies for the job profile,
• Standardize spelling variations,
• Avoid typical pitfalls.

Technically, the advisor could be implemented as a recommender system. After an initial learning phase in which competencies and their relationships are captured, the system can then independently generate individual suggestions. The goal is to increase user-friendliness while simultaneously ensuring the completeness and correctness of the data.

#Intelligently Clustering Competencies

A particular challenge lies in semantic standardization. Terms like “MS-Word” and “Microsoft Office” should be recognized as related - even though they have a part-whole relationship. Likewise, we want to systematically group thematically related competencies, e.g., in the area of agile software development.

For this, we tested various modern Natural Language Processing (NLP) methods for semantic clustering. One approach is based on embeddings: we convert competency terms into high-dimensional vectors using pre-trained transformer models (e.g., sentence-transformers). These capture semantic similarities, after which algorithms can identify dense clusters. Subsequent hierarchical clustering reveals the relationships between these clusters.

In further approaches, we transformed competency terms into hierarchical knowledge graphs using Large Language Models. The nodes represent individual competencies and the edges represent semantic relations between them. This creates a structured, ontology-like network of competencies that can serve as a basis for the recommendation functions.

#Maintaining Data Quality - Despite Free-Text Input

Another goal is the balance between flexibility and structure: users should be able to enter competencies freely without compromising data quality. Here, we are evaluating the use of a Large Language Model (LLM) for harmonizing inputs.

A language model analyzes entered free text and standardizes competencies unknown to the system in real time. The challenge is that the model must reliably distinguish between true synonyms (“JavaScript” and “JS”) and different concepts (“React” vs “React Native”). Through a confidence scoring system and human-in-the-loop, the model’s reliability can be improved. A continuous active learning system improves model accuracy through user feedback. Already learned synonyms are stored in the system and can be used for future entries.

This allows us to recognize and consolidate variants of the same term while avoiding the incorrect merging of different competencies.

The goal is to strike a balance between free data entry and a uniform data structure. Purely cluster-name-based input would ensure consistency but would severely limit users’ expressiveness. Conversely, complete freedom quickly leads to unwieldy and inconsistent data.

A possible middle ground: through early, AI-powered recommendations - for example, with the help of an LLM - inputs can be standardized without imposing rigid requirements. This maintains data quality while preserving user-friendliness.

#Finding Solutions Together - with Strong Partners

In our collaboration with the experts from the Mittelstand-Digital Zentrum Augsburg and AI specialists Dr. Martin Gottwald and Alexandros Tsakpinis (fortiss), we were able to validate our existing ideas and concretize new approaches for Profilery. Particularly valuable was the discussion of strategic questions such as:

• What do we specifically use AI for, and what added value do we want to create?
• What alternative approaches exist to solve a given problem?
• What data do we have available - and is it suitable for the respective AI approach?
• How do we objectively and traceably measure the success of our AI solutions?

Thanks to the intensive exchange, we were able to further develop our approaches - both technically and in terms of the concrete benefits for our users and customers.

#Conclusion

With the prototypically developed AI features, we are not only creating a foundation for the further expansion of our efficient profile management but also new opportunities for data-driven personnel decisions. Planned is the extension with predictive analytics for skill trends and an automatic matching engine that suggests the best-suited candidates based on project requirements.

The close collaboration with research partners shows: through the exchange between practice and science, it is possible to create AI solutions that are not only technically exciting but also effective in everyday business.