Peer Reviewer: Simulate Academic Peer Review With AI Before You Submit

Submitting to Nature or ICML without running your paper past a peer reviewer first is like deploying to production without testing. Peer Reviewer constructs reviewer personas from Zotero libraries and gives your paper a credible pre-submission review.

TL;DR

Peer Reviewer uses a multi-agent system — a Deconstructor, Evaluator, and Synthesizer — to simulate the academic peer review process. It draws on a Zotero library of reference papers to construct credible reviewer personas before you submit to any venue.

The 10-Second Pitch

Multi-agent review pipeline — three specialized agents: Deconstructor (finds weaknesses), Evaluator (scores against venue standards), Synthesizer (writes the review)
Zotero integration — reads your library to construct reviewer personas with relevant expertise
Venue-specific review criteria — configurable against any conference or journal's review rubric
Gap analysis — identifies missing citations, related work gaps, and methodological concerns
Rebuttal drafting — generates anticipated reviewer objections and suggested responses

Setup in 3 Steps

1. Install and connect Zotero: Set your Zotero API key and library ID in ~/.peer-reviewer/config.yaml.

2. Configure the venue:

yaml

venue: ICML 2026
review_criteria:
  - novelty
  - technical soundness
  - clarity
  - reproducibility

3. Run the review:

peer-reviewer review --paper ./paper.pdf --venue ICML --output review-draft.md

Prompt to test it:

Review this paper draft against ICML standards and give me a structured review with strengths, weaknesses, and specific suggestions.

How the Multi-Agent System Works

The three agents work in sequence:

1. Deconstructor Agent — reads the paper and your Zotero library, identifies the top 5 most relevant reference papers, and extracts what they do better or differently.

2. Evaluator Agent — scores the paper against venue criteria using a rubric. Identifies which claims are well-supported vs. speculative.

3. Synthesizer Agent — produces a formatted review that reads like a real reviewer report.

Pros / Cons

Pros	Cons
Surfaces weaknesses before reviewers do	LLM-written reviews lack expert human depth
Multi-agent architecture is credible	Zotero library quality determines reviewer persona accuracy
Configurable against any venue	Some fabrication risk — verify citation claims independently

Verdict

This is a genuinely useful pre-submission tool for academic researchers. No one wants their paper rejected for fixable clarity issues or missing citations when a weekend of AI-assisted review would have caught them. Use it as a first pass before sending to colleagues — not as a substitute for human expert review.