Robert The Bruce
PeerMind compresses the 3–6 month scientific peer-review cycle into a ~90-second multi-agent conversation.
Core trick
Runs parallel reviewer agents on a LaTeX project, then applies fix-agent unified diffs and recompiles in a sandboxed latexmk.
Description
PeerMind compresses the 3–6 month scientific peer-review cycle into a ~90-second multi-agent conversation. You upload a LaTeX project (or arXiv ID, or PDF), pick a target venue (NeurIPS, ICML, ICLR, Nature, Science, arXiv, or a custom venue), and PeerMind runs six Claude Managed Agents in parallel — two reviewers, a literature scout, a code runner, a fix agent, and an orchestrator — to simulate the full peer review process. The orchestrator synthesizes a final verdict with Opus 4.7 extended thinking streamed live to the UI, Fix Agent produces unified diffs that are applied inline with one-click Yes/No confirms, the paper recompiles in a Docker-sandboxed latexmk, and you can export the patched project back to Overleaf or draft a venue-style rebuttal — all in the same session.
README excerpt
# PeerMind **Your paper's toughest reviewer. In 90 seconds.** PeerMind compresses the 3–6 month scientific peer-review cycle into a ~90-second multi-agent conversation. Upload a LaTeX project (or arXiv ID, or PDF), pick a target venue (NeurIPS, ICML, ICLR, Nature, Science, arXiv, or a custom venue), and PeerMind runs **six Claude Managed Agents** in parallel: two reviewers, a literature scout, a code runner, a fix agent, and an orchestrator that synthesizes the verdict with **Opus 4.7 extended thinking streamed live to the UI**. Apply the fix-agent patches inline with one-click Yes/No confirms, watch the paper recompile in a Docker-sandboxed `latexmk`, and export the patched project back to Overleaf as a zip — or draft a venue-style rebuttal with the co-pilot in the same breath. Built for the [Built with Opus 4.7 × Cerebral Valley](https://cerebralvalley.ai/e/built-with-4-7-hackathon) hackathon. ## The three things that make this feel different 1. **Opus 4.7 extended thinking, exposed as a UI surface.** When synthesis runs, the reasoning tokens stream into a collapsible "Reasoning trace" bubble in the conversation rail — you literally watch Opus 4.7 weigh consensus issues, reconcile reviewer disagreement, and arrive at a calibrated acceptance-probability number. Not a log file. A first-class experience. 2. **Six Managed Agents, one narrator voice.** The right rail isn't a dashboard of cards — it's a conversation with a single persona (PeerMind) that narrates what each ag