Overview
arxiv-manager is an AI-powered assistant for creating challenging visual-reasoning Q&A tasks from scientific figures. It pulls papers from arXiv, extracts figures and context, and turns them into benchmark-style questions that demand precise visual interpretation.
Stack: Python · arXiv API · AI pipelines · PDF/figure extraction
Links: GitHub
Features
- arXiv ingestion — paper discovery and retrieval via the arXiv API, organized around focused fetch/query workflows
- Visual-reasoning task generation — composes Q&A items that require pixel-level and contextual reading of scientific figures
- Challenge-focused output — built for benchmark-grade tasks rather than surface-level captions
Design Philosophy
- Grounded in real papers — every task traces back to a source figure and its paper context
- Reproducible pipeline — scripted workflows for fetching, processing, and generating outputs
