cheap-scan1

What problem does it solve?

Cheap-scan1 dramatically reduces API token usage and cost when ingesting and analyzing academic PDFs by extracting text locally, triaging relevance, and only invoking vision models for pages that need them.

Core Features & Use Cases

• Local zero-token text extraction using pymupdf with per-page quality metrics to detect scanned PDFs and trigger fallbacks.
• Progressive triage funnel (abstract → introduction → conclusion → selective sections) and a --safe mode for exhaustive reads with per-section relevance tags.
• Pattern-driven scanner routing for tables, figures, and equations plus optional page rendering and parallel visual/equation analysis, producing a consolidated notes.md compatible with downstream literature workflows.
• Use cases: large-scale literature reviews, selective law-review processing, batch triage of working papers, and cost-conscious vision calls for empirical research.

Quick Start

Run the cheap-scan1 pipeline on path/to/paper.pdf in default mode to produce a consolidated notes.md in the specified output directory.