Three-stage methodology combining student surveys, LLM skill extraction, and dynamic preference-aware matching¶
Design · All designs
Evidence · 2 claims (2 for) · 1 study (1 design),q2· 0 of 1 report an effect size · 2 claims rest on one study
Description¶
The article proposes a data-driven team-formation framework with three stages: "student surveys to capture detailed and granular skill sets and project preferences; (2) LLM-driven extraction of essential project skills from sponsor-provided descriptions; and (3) a dynamic matching algorithm that iteratively prioritizes currently unfulfilled skills and integrates student preferences during team formation." It is positioned as combining bottom-up student agency with top-down project skill requirements in a single scalable pipeline, addressing gaps in tools like CATME Team-Maker that do not match skills to projects or dynamically integrate preferences.
Design Implications¶
Context¶
Requirements¶
- Students complete skill and preference surveys at the start of the course, with project summary descriptions available beforehand
- A skill list predetermined from prior and anticipated project requirements
- LLM-generated skill ratings are manually reviewed before use
Constraints¶
- Relies on self-reported skills and preferences, leaving it vulnerable to misrepresentation
- The preference weight parameter requires manual tuning
- LLM-generated project skills still require manual verification
Target Learners¶
- senior-level computer science capstone students
Learning Goals¶
- effective team formation for project-based learning
- student motivation and agency through preference alignment
Claims¶
- Dynamic Skill Matching Higher Skill Coverage [+M]
- Manual Assignment Highest Preference Satisfaction [+M]
Related Designs¶
- LLM-based project skill extraction workflow using a fine-tuned Llama 3.2 instruct model
- A three-stage design space for hybrid control in dynamic pairing: before, during, and after pairing up.
Examples¶
-
Key Sources¶
- Brandon Pardi, Garret Castro, Michael Pisman, Avash Adhikari, and Santosh Chandrasekhar. (2026). Improving Capstone Team Outcomes through Dynamic Skill Matching and Preference Alignment. https://arxiv.org/abs/2606.15572
Page metadata
| Field | Value |
|---|---|
type |
design |
id |
three-stage-llm-team-formation-methodology |
title |
Three-stage methodology combining student surveys, LLM skill extraction, and dynamic preference-aware matching |
description |
The article proposes a data-driven team-formation framework with three stages: "student surveys to capture detailed and granular skill sets and project preferences; (2) LLM-driven extraction of essential project skill... |
status |
review |
generated |
by process:settle-candidates on 2026-10-10 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — brandon-pardi-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.