Our AI Model

skillconomy goes far beyond the integration of external AI tools like ChatGPT. Our system is based on a proprietary, large-scale language model developed specifically for the requirements of the German-speaking job market and for use in active sourcing. The model is a central component of our platform and has been specifically trained to match profiles and requirements on a semantic level – taking into account information such as industry, company size, and, in particular, specific companies.

---

​

Differentiation from Other Providers

Many systems on the market rely on generic AI models trained on broad text corpora (e.g., Wikipedia, Common Crawl). These models are generally not specialized for the context of job profiles, career paths, or industry-specific requirements. Our architecture differs in the following ways:

• Proprietary model development: complete control over architecture, training, evaluation, and integration
• Specialized training focus: targeted training on German-language labor market and HR data
• Contextual knowledge about requirements, career paths, and specific companies
• Hosting and data storage in the EU in accordance with GDPR and the requirements of the EU AI Act

---

​

Company-Specific Context

A key distinguishing feature of our model is that it not only considers general industry characteristics or company sizes, but also integrates specific companies as contextual elements. This means: When our model sees a role, a skill, or a degree, it evaluates these not in isolation, but in conjunction with the employer – both for previous positions in the CV and for potential future roles.

​

Contextual Interpretation: An Example

The skill “Kubernetes” is a common requirement in technical roles – but its significance varies greatly depending on the company and task context:

• In a cloud-native tech company like “Google Cloud” or “Hetzner,” Kubernetes is typically a core component of productive system architectures. Here, the skill indicates deep experience in container orchestration, CI/CD pipelines, and infrastructure-as-code.
• At a mid-sized industrial company with a modern IT department (e.g., “Trumpf” or “Krones”), the same skill usually points to the ability to modernize existing on-prem or hybrid infrastructures and transition them into DevOps processes – often with a stronger focus on automation and stability.
• In a consulting role at companies like “Capgemini” or “Zühlke,” Kubernetes expertise often indicates breadth of project and technology experience – such as the ability to introduce Kubernetes into various client systems, document it, and transfer knowledge.

Our model recognizes these differences because it has learned from real data how specific companies evaluate, use, and embed skills into specific roles. This ability for contextual evaluation of individual terms – whether a skill, a job title, or a position – enables a deeper substantive classification and thus a significantly more precise matching compared to rule- or keyword-based methods.

---

​

Training Data and Model Size

The model was trained in three phases:

1. Pretraining on general German and English text sources (including Wikipedia, public documents)
2. Model-specific training on several million profiles from the German-speaking region in three stages of increasing difficulty

Training duration: >200 GPU hours on high-performance hardware
Number of parameters: several hundred million

---

​

Insights on Requirement Profiles

Through training on structured and unstructured HR data, the model has developed a detailed understanding of job requirements across various industries, company sizes – and specific employers. It recognizes, among other things:

• Which skills are typical or required for certain roles
• Which software solutions are commonly used in specific industries and roles
• Which industry experience is often required
• Which educational paths and qualifications are common for certain positions
• Which certifications are considered relevant (e.g., PMP, CFA, ITIL)
• In which roles language skills, willingness to travel, or leadership experience are required

---

​

Insights on Career Paths

In addition to analyzing requirement profiles, the model was trained to recognize typical career paths in the DACH region. It can identify relationships and patterns, such as:

• Which career steps typically follow one another (e.g., clerk → team leader → department head)
• From which industries or types of companies candidates move into certain roles
• Which universities or educational institutions correlate with certain employer groups
• How long specialists or managers typically stay in certain positions or companies
• Which positions serve as typical preparation for certain roles

Here, too, the company context is taken into account: A career move at a particular employer can – depending on industry, market position, and task structure – have very different meanings.

---

​

Data Protection & Compliance

All training data comes either from publicly accessible sources or has been contractually approved for use. Data processing takes place entirely within the EU. The models are hosted in certified data centers. Our system architecture is designed to meet the requirements of the GDPR as well as the upcoming EU AI Act – including traceability of model decisions (audit trails) and transparency regarding data sources.

---

​

Frequently Asked Questions

---

​

Contact

For technical inquiries or interest in more detailed documentation, you can reach us at any time at [email protected].