Built the backend for an AI tool that writes science test questions from lesson content.
- Two LLM agents: one writes questions, the other critiques them
- Led the prompt engineering and designed the quality scoring model
- Cut question writing from hours to minutes