A course has many data sources, such as lecture scripts, lecture recordings, homework assignments, slides, and the electronic semester apparatus of the library. Compiling these is initially a matter of document management: it must be simple and intuitive for instructors to compile all these documents in one place, which also meets the desire of students from a previous survey not to have to search for all materials at the beginning of each semester. At technical universities, many documents are based on LaTeX, which the system must directly understand.

The next step then becomes more interesting: Conversion of the documents into text form, standardization of the language, division into meaningful sections ("chunks"), and indexing of the same for easy findability. Finally, these chunks are made available to a LLM for processing inquiries. This, actually, is nothing new or particularly innovative; unique selling points here are only the coherence and automation of the document management and the special nature of our learning content.

These data sources are then drawn upon for the actual inquiries to the system, where it is important that the dialogue links back to the respective chunks of the original (document sections or time-​codes in videos). This allows one to trace and rework where the answers come from.