CSPaper
CSPaper provides an AI-driven platform that simulates the peer-review process for research papers before submission to target conferences or journals. The system analyzes uploaded PDFs or arXiv preprints against venue-specific rubrics, generating score-based feedback on dimensions like novelty, clarity, and significance in approximately 60 seconds. It supports 35+ venues including NeurIPS, ICML, ICLR, CVPR, and TMLR, and verifies related work citations to eliminate hallucinated references.
- Artificial Intelligence
- Software Only
- AI Agents
Funding
Founders
Founder details are not available yet.
Product
Problem
Academic peer review is a slow, inconsistent, and often opaque process that leaves researchers uncertain whether their manuscripts will meet the specific standards of target venues until after submission. The growing volume of research output has outpaced the capacity for thorough human review, leading to delayed decisions, subjective feedback, and missed opportunities for authors to address critical weaknesses before formal evaluation.
Solution
CSPaper provides an AI-driven verification and review platform that simulates the peer-review process for research papers before they are submitted to target conferences or journals. The system analyzes uploaded PDFs or arXiv preprints against venue-specific rubrics and templates, generating score-based, actionable feedback on dimensions such as novelty, clarity, significance, and correctness within approximately 60 seconds. Beyond surface-level assessment, CSPaper verifies the validity of cited related work, checks for potential desk-rejection factors like length and topic fit, and provides a prioritized list of improvements tailored to the chosen venue's standards. The platform is built on transparent, science-backed methodology, with review agents benchmarked against leading large language models to ensure reliability and rubric alignment. The system also supports arXiv LaTeX source code analysis and high-fidelity parsing of figures, tables, equations, algorithms, and references.
Target Audience
Primary users are academic researchers, PhD students, and industry research teams who submit papers to top-tier computer science conferences and journals and need rapid, rubric-aligned feedback to improve their manuscripts before peer review.
Features
- Venue-specific review agents that use official templates and rubrics for 35+ conferences and journals, including NeurIPS, ICML, ICLR, CVPR, and TMLR
- Verification of related work citations to eliminate hallucinated references, with explanations of why each suggested paper strengthens the submission
- Desk-rejection pre-assessment that flags issues with length, topic fit, quality, correctness, and prompt risks before full review
- Support for both PDF uploads and arXiv LaTeX source code, enabling analysis of figures, tables, equations, algorithms, and references with high-fidelity parsing
- Self-service review dashboard that allows users to inspect, sort, search, and delete past reviews and associated data at any time
- Iterative re-run capability that lets authors refine their papers and re-submit for updated feedback to track improvement over time
- Enterprise-level security agreements ensuring submitted papers remain confidential and are never used to train AI models
- Ongoing benchmarking of review agents against leading VLLMs including OpenAI and Google Gemini models to ensure reliability