In everyday school and university life, learners work on a large number of tasks, which can often only be evaluated to a limited extent. In particular, open task formats with explanations or complex solutions are time-consuming to correct, meaning that individual, constructive feedback is often not provided to the desired extent. At the same time, such re-registering students would be central to the learning process, as it not only makes the learning status visible, but also contributes specifically to further development. This is where the Chemie.KI project comes in. The aim is to develop an AI-supported assistance system that automatically evaluates pupils' and students' answers in the subject of chemistry. Answers are not only assessed in terms of their technical correctness, but also analysed in a differentiated manner according to predefined assessment criteria and provided with individual feedback. This is based on our specially trained AI models (>10,000 data sets, as of 03-2026), which are based on authentic student answers and didactically sound assessment logic.

Platform for teaching and research
Chemie.KI combines classroom application with empirical research. Teachers can create their own tests via a web-based interface, make them available using an access code and receive automated analyses of the learning status of their learning groups. Learners, in turn, receive structured, personalised feedback on their answers immediately after completing the test, regardless of time and place. At the same time, the anonymised data is used for research purposes. In this way, the underlying AI models can be continuously improved and systematically analysed with regard to their assessment quality. A central focus is on the question of the extent to which AI assessments agree with human assessors and under what conditions reliable use in the classroom is possible. Initial results show that trained models achieve a high level of agreement with human assessments and therefore have considerable potential for use in teaching-learning contexts.

Didactic perspective and objective
Chemie.KI aims to use the possibilities of digitalisation in a targeted manner to improve learning support. The combination of automated assessment, immediate feedback and data-based analysis of learning progress creates new opportunities for personalised learning and diagnostic insights into learning processes. Teachers are relieved of the burden and at the same time receive a sound basis for the adaptive further development of their lessons. In the long term, the system is designed to be flexibly expandable and can also be operated independently of external AI services in the future. Chemie.KI thus sees itself not only as a technical tool, but also as a contribution to the further development of examination and feedback cultures in science teaching.

Literature: T. Wilke, "Chemie.KI - Automatisierte Aufgabenbewertung durch trainierte KI-Modelle" , CHEMKON 2026, DOI: 10.1002/ckon.70016

www.chemie-ki.de