Documents are becoming structured data before they ever reach a human review screen. deepDoctection 1.2.x brings layout detection, table recognition, OCR, reading-order reconstruction, annotation linking, and structured export into one workflow, while DocStyle is connecting its document engine to AI assistants used by lawyers.
That combination points to a major shift in document processing: the page is no longer the final product. It becomes a source for organized text, figures, tables, relationships, provenance, and ordered information that software can analyze and export.
One Pipeline, Many Document Tasks
deepDoctection 1.2.x is built around a complete document analysis workflow. It combines layout detection with table structure recognition and OCR, then reconstructs reading order and links annotations before exporting the processed information in a structured form.
The framework can be configured with DocLayNet-based layout detection, Table Transformer structure recognition, and DocTR OCR. These components target different parts of the document, allowing the workflow to identify page elements, understand table structure, and extract text within one configured pipeline.
Its Page objects provide the central representation for the result. A Page object can represent text, figures, tables, relationships, provenance, and reading order, giving processed content a structure that extends beyond a flat OCR transcript.
That structure matters when a document contains more than paragraphs. Tables need their rows and columns preserved, figures need their place in the page captured, and relationships between annotations need to remain available for later processing. Reading order also gives extracted content a sequence that follows the document rather than the position of isolated text fragments.
From Model Configuration to Ordered JSONL
The sample code shows how a deepDoctection workflow can be assembled and inspected. It downloads a PDF named 2312.13560.pdf and an image named finance.png, creates the directories /content/docs and /content/imgs, and saves the files as /content/docs/paper.pdf and /content/imgs/finance.png.
The setup defines functions to show images and analyze files with deepDoctection. It also sets the environment variables DD_USE_TORCH to True, DPI to 200, LOG_LEVEL to INFO, and ENABLE_DYNAMIC_OBJECT_TYPES to False.
Dependencies installed by the sample include deepDoctection, transformers>=5.2.0, timm, python-doctr, pdfplumber, networkx, and lxml. The code retrieves the model profile Aryn/deformable-detr-DocLayNet/model.safetensors from ModelCatalog, reads its layout model categories, and registers the profile in ModelCatalog.
Configuration options control the analysis pipeline, including:
- USE_ROTATOR
- USE_LAYOUT
- USE_TABLE_SEGMENTATION
- USE_PDF_MINER
- USE_OCR
- USE_LAYOUT_LINK
- Weights for layout and item recognition models
The pipeline information is printed with IDs and names, and the meta annotation of the pipeline is printed as well. That inspection step exposes how the configured process is assembled, giving the workflow a visible record of its components and annotations.
deepDoctection also supports custom object types and custom PipelineComponent implementations. Those extensions can extract monetary and date entities, adding targeted information to the document representation instead of limiting the workflow to built-in page elements.
A custom pipeline can be assembled with ServiceFactory, filter, and service rollback. It can serialize processed pages and transform annotations into ordered JSONL chunks, creating an export path from analyzed documents to structured records.
Legal Document Tools Meet AI Assistants
DocStyle announced DocStyle AI on August 19, 2026, in Miami, FL. DocStyle is a provider of legal document repair, structuring, and automation solutions, and its new product brings that document engine into interfaces used by lawyers.
DocStyle AI connects the company’s document engine to Claude for Word, Microsoft Copilot, and a dedicated chat pane. The product is available for pre-order and will be generally available on October 1st, 2026.
The announcement describes the connection in direct terms: “The new solution connects DocStyle’s document engine to Claude for Word, Microsoft Copilot, and a dedicated chat pane, available for pre-order today ahead of general availability on October 1st.”
These developments occupy different layers of the document stack. deepDoctection focuses on turning pages into organized representations, with layout, tables, OCR, links, provenance, and reading order preserved through the process. DocStyle AI focuses on bringing a document engine into AI assistant experiences for legal work.
Together, they show where document intelligence is heading: extraction, structure, customization, and assistant access are becoming connected parts of the same story. As DocStyle AI moves toward its October 1st, 2026 general availability date and deepDoctection pipelines turn pages into ordered data, documents are gaining a new role—as information systems that software can understand, extend, and deliver through AI tools.
Based on
