AI-paper-reviewer

The review framework

The toolbox organizes its checks around a ten dimension framework that covers logic and argument, empirical rigor, writing quality, citations and attribution, math notation, double blind anonymity, IEEE formatting, language purity, structure and red flags. A four tier severity system marks each issue as Critical, Major, Minor or Pass so authors can prioritize. The prompts were assembled from multiple top conference reviewer guidelines, OpenReview discussions and the author's own submission experience.

How a self audit runs

An author opens a chat with any supported model, pastes a prompt such as 02_paragraph_audit, and then feeds the paper paragraph by paragraph. The model returns a strict review plus rewrite suggestions that can be applied directly. The repository ships prompt files for the full paper, the abstract, the introduction, the method, the experiments and the conclusion, plus specialized checks for double blind compliance, IEEE citation conversion, math notation consistency and an anti AI smell pass.

Reproducibility and rebuttal helpers

Beyond line by line editing, the project includes a reproducibility checklist assistant that walks through the twenty one item NeurIPS and ICML questionnaire in about a minute, and a double blind scanner that looks for real names, emails, school names, file paths and metadata left in the manuscript. A separate prompt helps draft a response to reviewer comments during the rebuttal phase. English users get a dedicated README and prompt set under prompts_en.

Editorial conclusion

AI paper reviewer is published under the MIT license and is written in Python. Its most used file is the paragraph audit prompt for steady revision passes.