AI Extract Section Options

Defines options controlling how AI Extract works for a Data Section.

Remarks

The 'AI Extract Section Options' class provides advanced configuration for how Grooper's AI Extract method interacts with a specific Data Section. These options allow you to control the alignment strategy used by the LLM (Large Language Model) when extracting section instances, ensuring that each section is accurately located and mapped back to the document for review, navigation, and validation.

Role in Grooper

• Used to customize the extraction and alignment behavior for individual Data Sections when using AI Extract.
• Supports both single-instance and multi-instance sections, adapting to a wide range of document layouts and data grouping scenarios.
• Enables integration of section-specific alignment instructions into the LLM prompt, improving extraction accuracy and reliability.

How It Works

• Each Data Section can be extended with an 'AI Extract Section Options' object, allowing you to override default alignment behavior.
• The 'Section Alignment' property determines what information is requested from the LLM (such as quotes, bounding boxes, or labels) and how Grooper interprets the LLM's output to identify section boundaries in the document.
• These settings are especially important for multi-instance sections, where accurate splitting and alignment of repeated content is required.

Configuration and Usage

• Choose an alignment mode that matches the structure of your documents and the nature of the section being extracted.
• For simple, single-instance sections, alignment may not be necessary. For repeating or complex sections, use quoting or geometric modes to improve accuracy.
• The selected mode impacts the JSON schema sent to the LLM, the expected response format, and the mechanism used to locate section instances in the document.

Example Scenario

For a document with multiple repeating claim sections, you might use 'StartQuote' to have the LLM return a quote from the first line of each section, allowing Grooper to split the document accordingly. For a section with a well-defined geometric region, use 'Geometric' to request explicit page and bounding box information.

These options provide granular control over AI-driven section extraction, supporting robust, accurate, and context-aware data capture in Grooper.

Properties

Name	Type		Description
Section Alignment	SectionAlignMode	►	Overrides the Default Section Alignment property of AI Extract for this section and its descendants. DefaultAuto Can be one of the following values: Name Value Description None 0 The section will not be aligned to the document. Only the section data is requested from the LLM; no alignment properties are included in the JSON schema. The section is used as-is, with no attempt to match or highlight it in the document. No navigation or highlighting will be available in the Data Grid or Data Viewer. Use this mode for single-instance sections where provenance is not important, or when alignment is not possible. Fastest extraction, but no review context or splitting for multi-instance sections. ► Auto 1 Infer the location of the section based on the positions of descendant fields. For multi-instance sections, Grooper infers the starting line of each section instance by analyzing the positions of descendant fields in the LLM response. The input is split accordingly, allowing for navigation and review of each section instance. For single-instance sections, behaves like 'None'. No additional properties are requested from the LLM; alignment is inferred post-extraction. Use this mode when section boundaries can be reliably determined from field positions, and explicit quotes or regions are not needed. ► Quote 2 Ask the LLM to quote the entire section content. The JSON schema requests a document_quote property from the LLM, containing a quote from the document with the full section content. Grooper uses this quote to locate and highlight the section in the OCR text. Useful for single-instance sections or when the entire section can be uniquely identified by its content. If the LLM returns an inaccurate or incomplete quote, alignment may fail. Use this mode when you want to ensure the exact document text for the section is captured and reviewable. ► StartQuote 3 Ask the LLM to quote the first line of each section instance, and then split the input at these positions. The JSON schema requests a begin_quote property from the LLM for each section instance, containing a quote from the first line. Grooper uses these quotes to split the document at the appropriate positions, enabling accurate extraction of multi-instance sections. Useful for repeating sections where each instance starts with a recognizable line or pattern. If the LLM returns inaccurate quotes, section splitting may be imprecise. Use this mode for multi-instance sections with clear starting lines. ► BoundingQuotes 4 Ask the LLM to quote the first and last lines of the section instance. The JSON schema requests both begin_quote and end_quote properties from the LLM for each section instance. Grooper uses these quotes to precisely identify the start and end of each section instance in the document. Provides the highest accuracy for complex or variable-length sections, or when both boundaries are needed to disambiguate instances. If either quote is missing or inaccurate, alignment may fall back to available information. Use this mode for multi-instance sections where both start and end lines are important for accurate extraction. ► Geometric 5 Ask the LLM to return a geometric region for the section. This mode must be used in conjunction with the Layout Objects quoting method, which provides the location of each text segment. The JSON schema requests __pageNo and __bounds properties from the LLM, specifying the page number and bounding box for the section. Grooper uses this information to directly extract and highlight the section region from the document geometry. Most useful when the LLM or extraction process can provide precise location data, such as for visually distinct or structured sections. If the LLM returns invalid or missing coordinates, alignment will fail. Use for advanced scenarios where spatial accuracy is critical, or when reviewing sections in visually complex documents. ► The 'Section Alignment' property determines how Grooper requests and interprets alignment information for this Data Section from the LLM. This setting controls what properties are included in the JSON schema, what the LLM is asked to return, and how section instances are mapped back to the document for navigation and review. For single-instance sections, alignment is often unnecessary. For multi-instance sections, alignment is critical for splitting and locating each instance. The selected mode impacts both the LLM prompt and the post-processing logic used to associate extracted data with document locations. Choose the mode that best matches your document's structure and the requirements of your extraction scenario.