Systems Group members Benjamin Hättasch, Leon Krüger and Carsten Binnig propose an approach to automatically derive suitable data schemas for relational databases based on the data that is inserted. This system called JUSTINE can reduce the overhead of using a relational database, since there is no need to manually craft a suitable schema. Furthermore, the schema can be automatically be updated when new kinds of data has to be stored. The system uses an automatic two-step mapping to find correct tables and columns to store the data in, and autonomously adjusts the schema were needed. JUSTINE was originally presented as a demo at VLDB'25. More information about JUSTINE can be found here: https://link.tuda.systems/JUSTINE

The poster was presented at the spring meeting of the database group of the German Society for Computer Science (GI). It took place in Regensburg on March 4th and 5th. The topic of this year's spring meeting was “Can We Trust the Model? Reproducibility and Reliability of LLM Results”. Read more about the event here.