Stile makes science lessons for schools, and teachers were spending hours writing test questions for them by hand. In a team of five, I built the Python and Flask backend for a tool that reads lesson content and generates curriculum-aligned questions automatically.
It uses two AI agents: one writes the questions and the other critiques them. I led the prompt engineering for both, tuning them to produce questions that match the curriculum and suit each year level, and proposed the scoring model that decides which questions are good enough to keep. Together, these turned a job that took hours into one that takes minutes.