Deeper Thinking Through Open Response: Using LLMs for Marking and Feedback

Sparx Science is an online homework system used weekly by 400,000+ students in England, part of the broader Sparx portfolio which serves 2,600+ schools and 2.2m+ students.

Traditionally, open-response questions—requiring students to answer with anything from a few words to full paragraphs—have required marking by teachers. As a result, online homework systems typically relied on constrained question types, such as multiple-choice.
However, Large language models (LLMs) have opened the door to reliably mark and provide meaningful feedback on open-response tasks, supporting effective retrieval practice and challenging students to think deeply.

In this talk, we'll share our approach to using LLMs to automatically mark prose in Sparx Science, including: fine-tuning, prompt engineering, and everything in between.