CAEDHET Ongoing Projects
Below you will find a selection of our current projects, which is continuously being updated:
ALMA
In collaboration with the ALMA project - Knowledge Networks in Medieval Romania of the Heidelberg Academy of Sciences, we are researching and testing the possibilities of AI in research and teaching. For example, we are evaluating the translation capabilities of large language models with regard to ancient languages such as Old French or Occitan and the possibilities of improving them through fine-tuning, as well as the understanding of historical technical terminology from the domains of medicine and law. Our attention is deliberately focused on all parts of the AI pipeline: from the development of new training data and supervised learning approaches to RAG architectures for the reproduction of factual knowledge. The systems created could later function as part of specialized agent systems that support users at various levels in the translation and development of historical text sources.
ASCVIT
The prototype of the AUTOMATIC STATISTICAL CALCULATION, VISUALIZATION AND INTERPRETATION TOOL was presented as part of a contribution to Towards Data Science. The tool, which is currently still based on Streamlit, enables the automatic, interactive interpretation of data sets. The visualizations are interpreted by combining statistical key figures from various methods with a specific configuration of the prompts, which are sent to an adapted LLM. The application currently enables descriptive analyses, hypothesis tests, regression analyses, time series analyses and cluster analyses. The code of the first version (V1) is available in the GitHub repository linked below. A publication is currently being prepared in the journal "Social Work, Media and Digitalization". Specifically, it shows how ASCVIT evaluates a fictitious data set from a social institution and can thereby provide important insights.
EDUMING
The EDUMING concept differs fundamentally from well-known approaches such as gamification, game-based learning, or serious games. While these concepts either integrate typical game elements into learning environments or use fully developed games for educational purposes, EDUMING pursues a constructivist approach in which the game is not only used but actively developed further. An EDUMING game is not a self-contained application, but rather a customizable template that can be modified and expanded using an integrated development environment (e.g., Unity or GameMaker Studio 2). Users decide for themselves whether the focus is on learning, playing, or both. The templates contain standardized learning mechanics that can be adapted or ignored. EDUMING promotes learning not only through playing itself, but especially through the creative development of game content – entirely in the spirit of Seymour Papert's constructionism. Supported by guidance and, if necessary, AI, players develop their digital skills, solve problems creatively, and create tangible artifacts that they can share and discuss in social contexts. The use of AI is intended to help automatically adapt existing game mechanics, specifically those related to learning, to the desired discipline. Atypical game genres such as survival games can also be used, as long as they contribute to the learning experience. EDUMING thus enables a flexible, practical, and future-oriented form of learning that not only imparts knowledge but also strengthens digital self-efficacy.
FACTS
FILTERING AND ANALYSIS OF CONTENT IN TEXTUAL SOURCES is an application for the autmatic conduction of meta-analyses. Configuring the system only requires a few steps. Depending on the topic or question, FACTS searches databases, extracts texts from articles and analyzes them using an LLM. The text is divided into chunks and the underlying topic structure of the question to be investigated is analyzed via LDA. The results are automatically interpreted and presented via interactive knowledge graphs. The functionality of FACTS will be presented at the IGPS Congress 2025 in Heidelberg. This is specifically about how AI can help answer current questions in school practice.
GAMMA
GAME AUTOMATED MECHANICS MODIFICATION & ADAPTATION is about the automatic adaptation of game mechanics by adapting the code of events in objects through an interactive interface. Currently, it is limited to the learning mechanics in EDUMING games that were developed using the GameMaker Studio 2 IDE. Depending on the topic and level, questions are generated by an LLM and implemented in the game. The aim of the application is, in keeping with the EDUMING concept, to simplify the adaptation and thus the development of digital learning games.
