Project-Based Learning: Chatbot Challenge
Published:
Some of the conversational AI systems I’m proudest of weren’t built by me at all — they were built by students I taught, in project-based courses I designed. Over the years I ran multiple chatbot courses at TU Berlin. Students built different kinds of systems:
- A tourism-information chatbot for Berlin
- Chatbot-based computer games
- Chatbots that train users to recognize the argumentation tactics of science denial and misinformation
The Challenge
Teaching conversational AI from a whiteboard is a poor substitute for building one. Students learn dialog design, evaluation, and the gap between a demo and a working system by shipping something real. Doing that well at scale meant designing a curriculum around project delivery rather than isolated assignments — coordinating teams of students, keeping technical scope realistic for a semester, and, and getting the content right.
My Approach
I ran each course as project-based learning: students worked in teams across a semester to design, build, and ship real chatbots rather than toy exercises. Each cohort had about 15 students and used technologies such as Python, large language models, Agentic AI, LangChain, Rasa, Docker, Vue.js, and Linux servers. I delivered the lectures and mentored teams directly through both build and evaluation.
Outcome
Students shipped working, evaluated chatbots rather than slideware — about 15 systems across the programs, several of which included real usability and user-experience studies. We also published two scientific papers on the courses:
- Die Chatbot-Challenge – Spielend mit KI von der Idee zum Dialogsystem
- Chatbots to strengthen democracy: An interdisciplinary seminar to train identifying argumentation techniques of science denial
Technologies: Agentic AI, Dynamic Prompting, LangChain, Rasa, Mistral LLM, Docker, Vue.js. Format: project-based learning.