Data Constellations is a multimedia project exploring differences in knowledge systems through classification of books. Ways in which books are classified in libraries have evolved over two and a half thousand years and have always been related with the changes in worldviews and in concepts of knowledge. Currently knowledge seems to be understood mainly as a network of data connections, what finds its expression both in bibliographic classifications and infamously biased AI training sets. The project explores the idea of knowledge as emerging out of interrelated data points. By turning different libraries’ descriptions of the same collection of books into self-organizing 3D objects, it makes abstract systems and concepts tangible. While encouraging appreciation of complexity of these structures, it also seeks to turn attention to their contingency on human subjectivity and interpretation.