paddleocr-doc-parsing

Multi-column PDF to Markdown conversion with layout preservation is achieved by running the layout-parsing script on a local or URL file. It returns structured Markdown and JSON outputs that retain headers, tables, formulas, and figures for direct pipeline ingestion.

PaddleOCR document parsing supports both local PDF files and remote image URLs as inputs. You provide the path to either source, and the skill processes the document to generate structured JSON and Markdown outputs while maintaining the original layout.

Extracting tables and formulas from invoices and research papers into JSON is supported through layout-aware parsing. The skill intelligently identifies and preserves these complex structural elements, outputting them within a complete JSON envelope for analysis and archival.

Before using PaddleOCR for document parsing, you need API credentials configured via environment variables. You also need to install dependencies including httpx, Pillow, and pypdfium2 to ensure the layout-parsing endpoint can process your PDFs and images correctly.

The best way to structure complex documents for knowledge base ingestion is using layout-aware parsing that outputs Markdown and JSON. This approach preserves multi-column relationships, headers, and figures, providing structured data ready for CMS or pipeline integration.

Document parsing output includes pruned results alongside Markdown text because the skill returns a complete envelope. This contains top-level text and per-page Markdown alongside pruned results to give you granular control over the extracted structured data.