BIBB Logo
An offer from the Federal Institute for Vocational Education and Training

How Aidy is becoming better and better

In this interview, Thomas Schmitz and Ruth Krumscheid explain the key aspects of the regular quality controls carried out on the “Anerkennung in Deutschland” chatbot.

Aidy, the AI chatbot at “Anerkennung in Deutschland”, supports skilled workers with questions relating to the recognition of their foreign professional or vocational qualification. The recognition chatbot operates nationwide and has conducted more than 50,000 chats since the start of 2025. Regular quality controls take place in the background to ensure that Aidy’s responses are always correct and helpful. In this interview, Ruth Krumscheid und Thomas Schmitz explain the key aspects of this process and talk about how Aidy is constantly improving.

When you monitor the chatbot, how can you tell whether Aidy’s answers are useful?

Ruth Krumscheid: Most chats proceed very well. Where necessary, Aidy poses important questions relating to matters such as place of work and profession or occupation. This develops the same sort of dialogue you would have with an actual person. Aidy is thus usually able to find the pathway to the appropriate information.

Thomas Schmitz: In rare cases, the answers may not be precisely the right fit, and you can normally recognise this from the course of the chat. A kind of “ping pong” will ensue. Skilled workers repeat their question several times and always receive the same or a very similar response. We look at such cases very carefully so that we can take corrective action and introduce improvement.

Ruth Krumscheid: Many users are now familiar with the idea of communicating with an AI-powered chatbot. If they fail to make progress, they will often simply start a new chat.

Why is quality control so important?

Thomas Schmitz: People don’t finish by saying: “Thanks Aidy, you’ve really helped me out.” At some point, the chat will just finish. This is why we examine a sample of the chats every week to check whether the responses were correct and helpful.

Ruth Krumscheid: Aidy is based on generative AI and draws on information from our quality-assured portal. But the chatbot formulates its own responses. This means that quality control is essential. Firstly, we are looking to see whether the mechanisms are working in the best possible way for Aidy and to identify where we might need to make adjustments. On the other hand, we receive feedback regarding the areas of the portal where there may still be room for improvement. We also ensure that the responses are consistent, i.e. that questions on the same topic are always answered in the same way.

Thomas Schmitz: Of course, recognition is a very complex subject. It is also intertwined with related issues such as entry to the country and looking for a job. Situations may occur in which Aidy’s answer is factually correct but does not address the objective of the skilled worker in question. We then take a close look and make a note of what went wrong and where.

How does the process from monitoring to improvement actually work?

Ruth Krumscheid: A small team of us meets up every week to look at areas in the chats which could be improved. We decide whether and which editorial or technical measures need to be adopted. The editorial team is responsible for making any changes to the pages on the portal. Sometime we stipulate answers for certain topics and write so-called “Aidy notes” which Aidy is then able to use.

Thomas Schmitz: Generally speaking, we always ask ourselves the following questions: To what extent should the chatbot be able to answer certain special questions? And when should Aidy be more proactive in pointing people towards other sources of information or guidance? The important thing is for the skilled workers to end up with a point of reference from which they can move forward.

What do you personally find exciting about working with Aidy?

Ruth Krumscheid: My main role is the technical management of the chatbot, and I find it really fascinating to see how Aidy processes and understands the information we provide. And to see sometimes how small adjustments can make a big difference.

Thomas Schmitz: The amount of linguistic development Aidy has undergone is also impressive. Aidy understands virtually every language. Responses are then given in German or English. Aidy can now also understand questions which are poorly phrased and can even react to keywords.

The interview took place in June 2026. Ruth Krumscheid is a specialist in technical applications, and Thomas Schmitz is an academic research assistant. They are based at the BIBB Centre of Excellence for the Recognition of Foreign Professional Qualifications, where they are jointly responsible for quality control of Aidy the chatbot. Contact for questions: Feedback on the portal

Please draw attention to Aidy or integrate it in your work in order to provide optimal support for skilled workers and to simplify the recognition process. You can try out Aidy here.