Unique Exam Simulator at the University of Bayreuth

Soon, students at the University of Bayreuth will be able to practise oral exams in a safe and controlled environment. In this simulation, students interact with a digital avatar that poses questions tailored to their specific field of study. These questions are AI-generated, ensuring unpredictability, and can also be adjusted to match the student’s skill level. The avatar is designed to adapt to the student’s responses, asking follow-up questions until the correct answer is provided. According to the ZHL, this oral exam simulator is the first of its kind worldwide.

“Preparing for oral exams requires mastering subject knowledge, managing situational stress, and communicating effectively. However, students often have limited opportunities to practise these scenarios and receive feedback on their performance. That’s why we developed this interactive AI-driven exam simulator,” explains Paul Dölle, research associate at the ZHL and project lead.

The simulation is recorded using a camera and microphone, and the performance is subsequently analysed with the help of AI. Students receive data-driven feedback on aspects such as eye contact, gestures, voice modulation, and the accuracy, thoroughness, and linguistic level of their answers. This data analysis is then discussed in a follow-up personal feedback session with instructors, helping to identify areas for improvement and enhancing the student’s exam performance.

“Our exam simulator can be used across disciplines, from humanities and natural sciences to law. Thanks to the adaptability of the AI, it can cater to specific examination requirements,” adds Dölle. The depth of the simulation depends on the input provided to the AI, such as lecture content, past exam questions, or study guides. Even the latest research can be incorporated into the simulation.

The AI integration is powered by a ChatGPT-based assistant, where lecture materials can be uploaded, and question formats can be programmed. The digital avatar is animated based on a video of the instructor, complete with a digitally cloned voice. After the practice session, audio and video recordings are evaluated using visual and audio analysis tools to assess body movements, speech, and tone. ChatGPT then analyses the spoken content based on the uploaded materials and provides detailed feedback. All recordings made during the simulation are stored locally on ZHL computers, ensuring that students retain full access to their data. Students can delete their recordings at any time, ensuring robust data privacy compliance.

The entire setup is being trialled during the 2024/25 winter semester with volunteers. However, students can already use the simulator for free in conjunction with a University of Bayreuth course.

The exam simulator was developed in collaboration with Tobias Hallmen, a research associate at the Chair of Human-Centred Artificial Intelligence at the University of Augsburg, and Dr Carl-Christian Fey, Managing Director of the DigiLLab at the University of Augsburg.