We strongly believe that AI-based chatbots greatly increase the possibilities for e-learning. This is because they combine the best of two types of learning: tutorial and constructivist. We are particularly interested in three use cases: Content-on-Demand, role-playing games, and free-text exercises. We have developed four examples of these that we would like to present to you. The first two are in German. This means that you can enter your input in English, but the chatbots will respond in German.
Content-on-demand (in German):
Cyber Security
Role-playing game (in German):
Sales conversation at a travel agency
Role-playing game (in English):
Conflict discussions
Free-text exercise (in English):
German Copyright Law
How we did it
All four examples are based on the general AI model Google Gemini 2.5 flash. We customized this model by training it with specific technical content (use case: information transfer) and adapting it with personas and behaviors (use cases: role play and free text task). The Chatbots usually play a more passive role. They answer users’ questions and react to their actions.
Use Case: Gathering information
In information gathering, customized chatbots can be used, for example, as a replacement for FAQs, for self-directed research, or as an agent that simplifies access to extensive information. We are thinking here, for example, of compliance issues and general training based on sets of rules. In theory, you can train chatbots with almost all types of data. Text-based information, like that commonly found in PDF files, is currently still particularly effective. Training with your own data ensures the technical quality of the chatbot, which cannot be guaranteed without training.
Use Case: Role-playing games
Role-playing games are a classic part of many behavioral training programs. In the e-learning context, these have only been implemented relatively poorly to date using simple simulations or branching elements. With AI chatbots, these role-playing games become much more realistic because the chatbots respond individually to the user’s input. “Customized” here means that we assign specific personas and scenarios to the chatbots and also provide them with criteria according to which the conversation should be evaluated.
Use Case: Free-text exercises
Free-text exercises in e-learning programs have always been challenging. The reason is that it’s almost impossible to give a satisfactory evaluation using the usual tools. Even small mistakes can lead to negative feedback. That’s why many e-learning designers avoid this type of task or get around the problem with predefined comparison answers. This means that learners must evaluate their own performance. With AI chatbots, it is now finally possible to automatically evaluate entries in free text fields.
Die Grenzen
AI chatbots are not meant to replace learning methods like tutorials, e-learning, role-playing in face-to-face training, or personal coaching. These supplements are a good addition.
There are also some limitations related to technology, organization, and data privacy. For companies, this usually means that
- A policy on AI should permit the use of chatbots.
- Ideally, the company is already using one or more AI chatbots, such as CoPilot, ChatGPT, Mistral, Claude or Gemini. Even better if they are hosted in-house .
- The information you type into the chatbot is not stored. This means that third parties cannot use it for evaluation. This is in accordance with both the General Data Protection Regulation (GDPR) and the EU AI Act.
- Free-text exercises that use AI should not be used for exams. If they are, they will quickly go beyond the allowed limits.
- The use of AI is clearly stated.
Our use cases
Cy B. knows the answer
Cy B. is an German-language AI chatbot that answers your questions about cybersecurity.
With this prototype, we want to test the extent to which chatbots are suitable for knowledge transfer and consolidation. Cy B. was trained using publicly available information from the BSI (German Federal Office for Information Security) and ENISA (European Union Agency for Cybersecurity), but also uses other sources.
Start Chatbot
Please use a Chromium-based browser, such as Chrome, Edge, Brave, or DuckDuckGo. The AI model used is Google Gemini 2.5 Flash. The chatbot and the underlying Google Cloud infrastructure are operated in the U.S.
Please note that we are not liable for the accuracy of the chatbot’s statements. Technical errors are inherent to LLM-based chatbots. If you have any concerns, please consult cybersecurity specialists or the institutions mentioned above.
Leonie wants to go on vacation
In this German-language role-playing chatbot, users advise Leonie, a customer at a travel agency.
Sales and distribution training is important in almost every industry. With this chatbot, we demonstrate how AI can be used as a role-play partner in learning environments. After a successful or unsuccessful consultation, users receive an analysis of their conversation.
Start Chatbot
Please use a Chromium-based browser, such as Chrome, Edge, Brave, or DuckDuckGo. The AI model used is Google Gemini 2.5 Flash. The chatbot and the underlying Google Cloud infrastructure are operated in the United States.
We have not specifically trained the chatbot. Our role here is primarily to develop a persona with credible behavior patterns and define evaluation criteria for assessment. The “travel agency” example can be replaced by any other industry.
Tanja is challenging
This chatbot allows aspiring filmmakers to practice difficult conversations.
This chatbot is an English-language role-playing game. Users can choose from three scenarios typical of the film industry. Acting as line producers, users are responsible for the economic and organizational aspects of a film. They must collaborate with Director Tom to find solutions to the scenarios. After ending the role-playing game with the prompt “End,” users receive detailed feedback on their conversation skills.
Start Chatbot
Please use a Chromium-based browser, such as Chrome, Edge, Brave, or DuckDuckGo. The AI model used is Google Gemini 2.5 Flash. The chatbot and the underlying Google Cloud infrastructure are operated in the United States.
We have not specifically trained the chatbot. Our role here is primarily to develop a persona with credible behavior patterns and define evaluation criteria for assessment.
Free-text exercise German Copyright Law
In this free-text exercise, learners are asked to evaluate a case study.
Case studies are the ideal tool for bridging the gap between theory and practice. Therefore, exercises that use case studies as a starting point are perfect for knowledge transfer. Our example addresses a common problem in the daily lives of filmmakers: Should I reveal my idea or not? The learners’ answers are analyzed and evaluated by an AI chatbot working in the background.
Start Exercise
Please use a Chromium-based browser, such as Chrome, Edge, Brave, or DuckDuckGo. The AI model used is Google Gemini 2.5 Flash. The chatbot and the underlying Google Cloud infrastructure are operated in the United States.
We have not specifically trained the chatbot. Our role here is primarily to define which statements are desired and to establish evaluation criteria for the analysis. The example “copyright” can be replaced by any other industry.