NeuroDong/Ai-Review

What it solves

Researchers often struggle to identify the strengths and weaknesses of their manuscripts before submission. Ai-Review provides an automated way to get structured feedback, evaluate the quality of figures and layout, and determine how a paper's competitiveness compares to other works in the same field.

How it works

The system uses a combination of Large Language Models (LLMs) and Vision-Language Models (VLMs) to analyze manuscripts in PDF, LaTeX, or Word formats. It employs advanced prompt engineering techniques—including Chain-of-Thought, Few-Shot, and Reverse Prompting—to generate detailed reviews. For visual analysis, it captures page snapshots to evaluate layout and image aesthetics. It also includes a specialized tool to detect prompt-injection attacks hidden within PDFs to ensure review integrity.

Who it’s for

Academic researchers and authors who want to improve the quality and acceptance probability of their scientific papers.

Highlights

• Relative Rank: Ranks a manuscript against a pool of papers from target venues (e.g., ICLR, NeurIPS) to gauge competitiveness.
• VLM-Powered Review: Uses vision models to analyze figures, equations, and overall page layout.
• Agent Integration: Can be installed as a "Skill" on agent platforms like Cursor to review files directly in the chat.
• Prompt Injection Detection: Scans PDFs for hidden instructions designed to manipulate AI reviewers.
• Structured Feedback: Generates specific lists of strengths, weaknesses, and improvement suggestions.