Answering student questions from the university’s own documents
Students ask the same procedural questions; the answers are buried in PDFs.
The problem
Students at the University of Antwerp keep asking the same procedural questions about thesis rules, internships and study-guide details. The answers are already written down, just scattered across long PDFs that few people read end to end.
The approach
I built a retrieval-augmented assistant that answers strictly from the official documents. It splits the source files into passages, embeds them in a vector store, retrieves the passages most relevant to each question, and asks a language model to answer only from that retrieved context, citing the source and declining when the documents don’t cover it. The prompt is hardened against attempts to override those rules, and every question and answer is logged for review.