JoyDefense Multi-Advisor Thesis Review Coordinator

Problem

Pre-defense thesis review often depends on a single reviewer perspective: one person focuses on statistical rigor, another on writing clarity, and another on novelty. Asking multiple advisors in sequence can make feedback look homogeneous and make it hard to decide which revisions matter most before an actual committee session. JoyDefense makes this workflow explicit: it parses a thesis into structured information, then drives advisors with different persona settings in parallel, and finally synthesizes consensus, disagreements, and priority revisions.

How It Works

The core flow has three stages.

• Thesis parsing: extract information from title, abstract, research problem, innovation, methodology, results, figures/tables, writing quality, and references, giving the review step a shared input.
• Parallel advisor review: each advisor responds across Judgment, Management, and Persona, covering the reviewer’s assessment, process concerns, and questioning style. Registered advisors can be aligned to different emphases, such as methodology/experimental design, novelty, and writing/structure.
• Synthesis: aggregate advisor opinions, consensus points, divergent points, and the highest-priority revision items, helping the author address the most committee-risk issues first.

The skill also exposes advisor-management commands such as /add-advisor, /remove-advisor, /list-advisors, and /update-advisor. When transcribed audio and thesis materials are available, the update path can refine a single advisor’s review preferences. Each update targets one advisor and keeps a backup in versions/.

Boundaries And Notes

JoyDefense is better used as a structured pre-defense review scaffold than as an automatic fact-checking tool. Its output depends on thesis parsing quality: if the abstract, method, or results are sparse, advisor feedback may remain surface-level. Advisor persona settings are configuration-driven, and the default list is closer to example roles such as Zhang, Li, and Wang. Before formal use, align the advisor code, research direction, and review emphasis so the output style does not drift from the real committee. For domain-sensitive judgments, still review manually against advisor recordings, target thesis, and field conventions.