Leben in Deutschland: An Intelligent Tutoring Game for the German Citizenship Test

"Leben in Deutschland" is an intelligent tutoring game designed to help learners prepare for the German citizenship test, which requires answering at least 17 of 33 questions correctly drawn from a 310-question database. The work proposes pairing a serious game set in a German town environment with a locally run LLM tutor. The tutor, built on Llama 3.2 3B and a retrieval-augmented generation system drawing on roughly 758,000 words of civic-education material, tracks mastery of individual questions, explains failed answers, and responds to open-ended queries.

The central finding is that LLM-generated personalized feedback appears most valuable for open-ended questions that cannot be produced through conventional authoring, while explanations for standardized multiple-choice items could be written manually or drafted with LLM assistance. Retrieval failures sometimes led to generic or off-target explanations, underscoring the dependence of response quality on the retrieval pipeline.

Evidence comes from a mixed-methods pilot with only four participants. Reported ratings were high across learning effectiveness (4.8 of 5), game design (4.6), and LLM performance (4.0), with three of four participants finding the feedback helpful. These results should be treated as preliminary: the sample is very small, the questionnaire was not validated, exposure time was short, and some question wording was suggestive.

The authors also identify the game's simple design as a possible constraint on long-term engagement. Planned future work includes migrating to a Godot-based implementation, expanding the question database, adding non-player characters and interactive objects, and conducting a larger evaluation of learning outcomes with actual citizenship test candidates.