PaperMole: LLM-assisted data extraction and quality appraisal for systematic reviews

PaperMole automates data extraction and quality appraisal for systematic reviews. Given a corpus of research papers as PDFs, it produces a structured spreadsheet with one row per study and user-defined extraction fields such as sample size, study design, or reported performance metrics. The pipeline parses PDFs including tables and figures, uses a large language model to extract the requested fields, scores each study against standard reporting checklists (PRISMA, CLAIM, QUADAS-2), and exports the result as a spreadsheet. Every extracted value carries two provenance annotations: a certainty label (high, moderate, low, not_reported, conflicting) and the verbatim quote from the source document that the value was drawn from. This makes each extraction auditable rather than opaque. Intended use: PaperMole produces a first draft for human verification, not a finished extraction. Large language models make errors, and values flagged low or conflicting in particular require manual checking before use in a published review.