Docling Parser

The Docling parser is a third-party intelligent document parsing framework that brings together multiple open-source models for layout understanding and table extraction. To enhance responses from information available in images and complex tables, combine the Docling parser with image processing by using Large Language Models (LLMs).

Note Using LLMs to parse images impacts credit consumption and is billed against the Intelligent Processing usage type. For more information, see Billing Considerations for Unstructured Data.

With LLM-based image processing, Data 360 filters out unnecessary images such as logos and routes only complex infographics. Data 360 captures and sends pages containing more than a specific threshold of images as a single whole-page image for captioning. This captures any relationships between multiple images, maintains quality, and reduces latency.

With LLM-based image processing, Data 360 evaluates and routes only complex tables to the LLMs. These include tables with nested headers or long-form merged columns that span multiple pages.

For more information, see Docling parser.

Use the Docling parser to parse content that contains information in these formats.

Parser Option	Format	Description	Extension
Docling	Document	Structured content such as text and tables in documents.	.pdf, .docx, .doc
	Spreadsheet	Spreadsheet data in cells, rows, and columns.	.xlsx, .xls
	Slide Deck	Content in presentations such as PowerPoint.	.pptx, .ppt
	Web Page	Content in websites.	.html, .xml, .xhtml, .htm
Docling + LLM-Based Image Processing	Image File	Content in images, diagrams, and flowcharts. Parse images using LLMs by turning-on the image-processing toggle when you define chunking strategies.	.png, .jpg, .jpeg, .bmp, .tiff, .tif, .gif, .pdf
Docling + LLM-Based Image Processing	Complex Tables	Content in tables with merged cells or columns and nested multi-level headers. Parse complex tables using LLMs by turning on the image-processing toggle when you define chunking strategies.	.pdf

You can’t select both the Docling parser and LLM-based Visual Data Pre-processing options for a search index.

To parse images using an LLM, turn on image processing when you select chunking strategies. For more information on configuring content parsing with Data 360 search indexes, see Create a Vector Search Index with Advanced Setup and Create a Hybrid Search Index with Advanced Setup.

Example Use Case: Select Docling parser with image processing in scenarios where business-critical information is embedded within visual elements of enterprise documents.

Consider an airline maintenance division that manages operational processes through structured RACI-style flowcharts within its governance manuals. These process maps define critical handoffs and validations across functions such as operators, reps, modules, planning, and operations. Each diagram includes directional arrows, decision nodes, and annotated swim lanes, reflecting step-by-step responsibilities from irregularity detection through to work plan reprioritization and resource redistribution.

With this enhancement, the LLM can interpret the flowchart as a structured process, recognizing nodes, connectors, arrows, and step order. Instead of a flat list of labels, the LLM reconstructs the sequential flow of actions across roles and departments. As a result, answers to questions from the field related to responsibilities or the sequence of steps are more accurate.

Example Use Case 2: If business-related critical information is embedded in complex tables, select the Docling parser with image processing. This Accounts Management table contains nested headers and rows of dense text.

Data 360 chunks the information in the table and preserves the header information before sending the chunks to the LLM. This process improves response accuracy and reduces hallucinations.

A sample accounts management table with nested headers and rows with dense text.