Abstract
Open science initiatives seek to make research outputs more transparent,accessible, and reusable, but ensuring that published findings can beindependently reproduced remains a persistent challenge. This paper introducesOpenPub, an AI-powered platform that supports researchers, reviewers, andreaders through a suite of modular copilots focused on key open science tasks.In this work, we present the Reproducibility Copilot, which analyzesmanuscripts, code, and supplementary materials to generate structured JupyterNotebooks and recommendations aimed at facilitating computational, or "rote",reproducibility. We conducted feasibility tests using previously studiedresearch papers with known reproducibility benchmarks. Results indicate thatOpenPub can substantially reduce reproduction time - from over 30 hours toabout 1 hour - while achieving high coverage of figures, tables, and resultssuitable for computational reproduction. The system systematically detectsbarriers to reproducibility, including missing hyperparameters, undocumentedpreprocessing steps, and incomplete or inaccessible datasets. Whilepreliminary, these findings suggest that AI-driven tools can meaningfullyreduce the burden of reproducibility efforts and contribute to more transparentand verifiable scientific communication. The modular copilot architecture alsoprovides a foundation for extending AI assistance to additional open scienceobjectives beyond reproducibility.