Using Large Language Models to Support High Volume Application Review for an Undergraduate Research Program

In three linesPurdue University deploys GPT-4o, GPT-5-mini, and GPT-5.2 to evaluate 1,200 applications for the SURF 2026 program. Models score statements of purpose across 6 rubric categories (0-3 scale), generating scores and rationales in 4.6 hours. GPT-5.2 shows strongest rubric adherence. Final coordinator review takes 4 hours versus multi-week effort in prior cycles.

Summary generated by Claude — human-verified