Docling

Accepts PDFs, DOCX, PPTX, images, and HTML as input and routes each through format-specific parsers before converting to a unified internal document representation. Uses format detection to select appropriate extraction engines (e.g., pdfplumber or pypdf for PDFs, python-docx for DOCX, PIL for images), normalizing all outputs into a common DoclingDocument AST that preserves structural metadata.

Analyzes spatial positioning, bounding boxes, and visual hierarchy of document elements (text blocks, tables, images, headers) to reconstruct logical reading order and document structure. Uses computer vision techniques to detect page regions, classify element types by position and styling, and build a hierarchical representation that preserves the original layout semantics rather than flattening to linear text.

Automatically detects the language of document content and applies language-specific processing (OCR language models, text segmentation, heading detection) appropriate to the detected language. Supports 50+ languages including CJK, Arabic, Devanagari, and Latin scripts, with configurable language hints for ambiguous cases. Preserves language information in document metadata for downstream processing.

Processes large documents (>100 MB) in a streaming fashion, parsing pages or sections incrementally rather than loading the entire document into memory. Yields DoclingDocument chunks as they are processed, enabling memory-efficient handling of very large files and progressive output generation without waiting for complete document processing.

Splits extracted document structure into chunks suitable for RAG systems, respecting semantic boundaries (paragraphs, sections, tables) rather than naive character-count splitting. Implements configurable chunk size, overlap, and boundary detection to preserve semantic coherence while enabling efficient retrieval. Maintains chunk metadata (source page, section, confidence) for traceability.

Detects table regions within documents using visual boundary detection and extracts cell contents while maintaining row/column relationships. Handles merged cells, multi-line cell content, and nested tables by parsing table structure into a normalized grid representation with explicit row and column indices, then exports to structured formats (JSON, Markdown table syntax) that preserve cell boundaries and relationships.

Detects when documents contain image-only content (scanned PDFs, photographs) and automatically routes them through an OCR engine (Tesseract, EasyOCR, or cloud-based APIs) to extract text. Preserves spatial positioning of recognized text by mapping OCR bounding boxes back to document coordinates, enabling layout analysis and table extraction to work on scanned documents with minimal quality loss.

Converts DoclingDocument AST to Markdown format, mapping document structure (headings, lists, tables, emphasis) to Markdown syntax while preserving hierarchical relationships. Uses the layout analysis output to infer heading levels from visual hierarchy, converts table structures to Markdown table syntax, and preserves inline formatting (bold, italic, links) from source documents.

Exports DoclingDocument to JSON format with complete metadata including bounding boxes, element types, confidence scores, and hierarchical relationships. Each element (text block, table, image, heading) is represented as a JSON object with spatial coordinates (page number, x, y, width, height), content, and type classification, enabling downstream systems to reconstruct document layout or perform spatial queries.

Processes multiple documents in sequence or parallel, with configurable batch size, timeout handling, and progress reporting. Implements error handling per-document so that failures in one document don't halt the entire batch, and provides callbacks or logging for monitoring processing status, memory usage, and performance metrics across the batch.

Splits documents into chunks optimized for RAG systems by respecting document structure (chapters, sections, paragraphs) rather than naive character-count splitting. Uses layout analysis to identify logical boundaries (heading changes, section breaks) and creates chunks that preserve semantic coherence, with configurable chunk size, overlap, and metadata preservation (source page, section title, heading hierarchy).

Allows users to define custom classification rules or provide trained models to tag document elements (text blocks, tables, images) with domain-specific labels (e.g., 'disclaimer', 'product-spec', 'pricing-table'). Integrates with the layout analysis pipeline to apply classifiers to detected elements and attach custom tags to the DoclingDocument AST, enabling downstream filtering or specialized processing based on element type.

Compares two DoclingDocument objects to identify structural and content differences, including added/removed elements, modified text, table changes, and layout shifts. Produces a diff report showing which elements changed, their locations, and the nature of changes (content modification, structural reorganization, element addition/deletion), useful for version control or change tracking in document processing pipelines.