Bachelor- & Masterprojekte entdecken

Developing Unplugged AI Learning Materials for K-12 Students: A Competency-Based Approach

This thesis investigates key AI competencies that K-12 students should acquire, such as algorithmic thinking, pattern recognition, and ethical awareness [1-3]. Based on these competencies, unplugged (non-digital) learning activities will be designed to engage young learners in foundational AI concepts [4,5].

[1] https://dl.acm.org/doi/full/10.1145/3685680
[2] https://dl.acm.org/doi/10.1145/3313831.3376727
[3] https://ojs.aaai.org/index.php/AAAI/article/view/5053
[4] https://www.aiunplugged.org/
[5] https://www.i-am.ai/de/build-your-own-ai.html

Enhancing Pre-Service Teachers’ Explanations of Computer Science Concepts through AI-Agent Feedback

This thesis investigates how a visual or acoustic AI-agent system (such as D-ID’s AI Agents) can be used in experiments with pre-service teachers to improve their explanations of K-12 computer science content. Building on research about professional knowledge and explaining skills [1-3], the study engages pre-service teachers in explaining selected CS topics (e.g. algorithms, binary numbers, or AI basics), receiving AI-driven feedback, and iteratively refining their explanations. The goal is to evaluate whether interaction with the AI system supports the development of clearer, more structured and pedagogically effective explanations.

[1] Kulgemeyer, C., et al. Professional knowledge affects action-related skills: The development of preservice physics teachers’ explaining skills during a field experience. DOI: 10.1002/tea.21632
[2] Findeisen, S., Deutscher, V. K., & Seifried, J. Fostering prospective teachers’ explaining skills during university education — Evaluation of a training module. DOI: 10.1007/s10734-020-00601-7
[3] https://link.springer.com/article/10.1007/s10758-025-09875-1
Using: D-ID AI Agents

Click & Predict - Exploring Autoregressive Text Generation with Interactive LLM Visualizations

This thesis designs an interactive visualization tool (e.g. dendrogram or branching tree) that allows K-12 students to manually sample the next token to explore how large language models generate text autoregressively. Using tokenization examples from https://tiktokenizer.vercel.app, students can click to select the next token at each step, observing branching probabilities and the sequential process of text generation. The tool aims to make abstract concepts like probabilistic token selection and autoregressive generation tangible and engaging for young learners.

[1] https://tiktokenizer.vercel.app

Developing an AI-Driven Classroom Simulator: Integrating Automatic Speech Recognition, Language Models, and Text to Speech models for Educational Practice

Recent advances in automatic speech recognition, large language models (LLMs), and text-to-speech synthesis have opened new possibilities for creating interactive and realistic educational simulations. This master's thesis explores the development of a simulated classroom environment populated by artificial student agents. These agents are capable of voice-based interaction using state-of-the-art automatic speech recognition (ASR), generative LLMs for dialogue, and expressive text-to-speech (TTS) systems. The primary goal is to identify and integrate the most effective combination of models to simulate realistic classroom conversations, questions, and misunderstandings. Building on recent research in AI-based tutors and synthetic student modeling the thesis includes a benchmarking study of various ASR, LLM, and TTS systems in educational contexts, an implementation of a small-scale simulation prototype, and an evaluation of the system's realism and educational value. The work has implications for teacher training, educational software testing, and research on human-AI interaction in learning environments.

Literature: https://www.sciencedirect.com/science/article/pii/S1096751624000526

Contact

Nach oben scrollen