AI assisted marking and feedback

Academics at the University already spend disproportionate amounts of time marking students’ assignments. As staff to student ratios threatened to worsen, we wanted to explore how AI could ensure that teaching workloads remain sustainable.

Delivery

Using OpenAI’s Assistants API, we were able to upload student reports, marking criteria, and model responses to an LLM, and have it use these resources to generate feedback. The tool completed in minutes work that would have taken a manual marker an estimated 18 hours, and produced uniform feedback across submissions.

Outcome

Through extensive iteration, we uncovered powerful workflows for enabling complex, LLM-driven tasks – e.g. by supporting non-technical users to construct machine-friendly prompts and leveraging RAG patterns to derive meaningful LLM outputs. The result is an effective essay marking tool that broaches big questions around the role of AI at the University, and a set of technical blueprints for our future AI workflows.