Back to Research
RSMS: Adaptive Research Student Mentoring via SRL-Aligned and Contestable AI Feedback
RSMS is a multi-agent mentoring system designed to support self-regulated learning in research students through SRL-aligned weekly feedback and a contestable AI mechanism.
CompletedStarted: Oct 4, 2025Category: Education AI
Multi-Agent SystemsContestable AILLM-as-JudgeSelf-Regulated LearningHuman-in-the-Loop AI
Research Thumbnail
Research Objectives
- •Deliver SRL-aligned weekly mentoring feedback covering forethought, performance, and self-reflection
- •Enable contestable AI feedback through evidence-based student justifications
- •Encode supervisor resolutions into adaptive memory rules for future feedback cycles
- •Evaluate mentoring quality via LLM-as-Judge rubrics with inter-rater reliability checks
- •Maintain active deployment for real research students at Ozlab, Nagoya Institute of Technology
Methodology
Grounded in Zimmerman's three-phase self-regulated learning model, the system enforces coverage of forethought, performance, and self-reflection through a structured output schema. The core contribution is an Alignment Agent that enables evidence-based feedback challenges, with supervisor-in-the-loop adaptive memory that stores resolved cases as reusable guidance rules.