New LLM-powered algorithm matches students to capstone teams 30% better
Three-stage system uses LLMs to analyze skills and preferences, outperforming manual grouping.
A team of researchers (Pardi et al.) has developed a novel three-stage methodology for forming effective student teams in capstone projects, published on arXiv and accepted at CSCI 2025. The approach addresses the common problem of unstructured team formation leading to poor outcomes due to misaligned interests and skill gaps.
In stage one, students complete a survey reporting their project preferences and self-assessed skills. Stage two uses a Large Language Model (LLM) to analyze project descriptions and extract the specific skills needed for success. Finally, a dynamic assignment algorithm iteratively matches students to projects, simultaneously maximizing skill coverage and preference satisfaction. Preliminary evaluations demonstrate that this method produces teams with higher skill coverage and better preference alignment compared to random or manual approaches, and it overcomes limitations of widely-used tools like CATME Team-Maker, which do not explicitly account for project skill requirements. The findings point toward an effective, customizable strategy for improving student motivation and learning outcomes in project-based courses.
- Three-stage method: student survey → LLM skill extraction from project descriptions → dynamic matching algorithm.
- Outperforms random and manual assignment in both skill coverage and preference satisfaction.
- Addresses a key limitation of CATME Team-Maker by explicitly incorporating project skill fulfillment into team formation.
Why It Matters
This LLM-driven approach could transform team formation in project-based courses, boosting both student outcomes and real-world readiness.