Slug: llm-guidance-builder Version: 1.0.0 Infer: No Event Aware: Yes

​

Overview

​

LLM Guidance Builder Model

The LLM Guidance Builder automates the creation of extraction guidance from example documents. When you mark a document as “guidance,” this model analyzes it and generates extraction patterns that improve accuracy for similar documents in the future.

​

How It Works

1. Guidance Identification: Recognizes documents marked as “guidance” for processing
2. Data Extraction: Extracts structured data from the guidance document using the data definition (taxonomy)
3. Pattern Recognition: Maps the extracted data to its location in the document
4. Guidance Generation: Creates reusable guidance patterns based on successful extractions
5. Guidance Storage: Stores the guidance for automatic application to similar documents

​

Options Configuration

Option	Description
taxonomy	The data definition that specifies what information to extract
data_store	The data store where extracted data is stored
guidance_set	The set where generated guidance will be stored
build_guidance	When enabled, automatically builds guidance from documents with the “guidance” label
interactive_mode	When enabled, automatically reprocesses existing documents when guidance is updated

​

Message Templates

The model provides these contextual message templates to help you work with guidance:

• Suggest Data Definition: Get AI suggestions for a data definition based on document content
• Discuss Differences: Compare extracted data with external data to identify discrepancies
• Summarize: Generate a concise summary of selected content

​

Process Flow

​

Guidance Structure

The generated guidance contains:

• Document Reference: Links to the original guidance document
• Extracted Data: The structured data extracted from the document
• Location Mapping: Where in the document each data point was found
• Applicable Tags: Which data elements this guidance applies to
• Guidance Key: Optional identifier for controlling guidance application

​

Example Usage

The LLM Guidance Builder is particularly useful for:

• Creating Extraction Templates: Turn successful document extractions into reusable patterns
• Improving Extraction Accuracy: Apply lessons from one document to similar documents
• Interactive Learning: Continuously improve extraction as you process more documents
• Building Knowledge Bases: Create a library of extraction patterns for different document types
• Reducing Manual Configuration: Minimize the need to configure extraction rules manually

​

Workflow Example

1. Process a document with the Data Labeling model to extract data
2. Mark the document as “guidance” when extraction is accurate
3. Run the Guidance Builder to create reusable patterns
4. Process similar documents, which now benefit from the guidance
5. Iteratively improve guidance by adding more example documents

​

Interactive Mode

When interactive mode is enabled, the model:

• Monitors for document updates that affect guidance
• Automatically reprocesses existing documents when guidance changes
• Creates a feedback loop that continuously improves extraction quality

---

​

Inference Options

The following options can be configured when using this model for inference:

Name	Label	Type	Description	Default	Required
taxonomy	Data Definition	taxonomy	The data definition to use for the model	-	No
data_store	Data Store	tableStore	N/A	-	No
guidance_set	Guidance Set	guidanceSet	N/A	-	No
build_guidance	Build Guidance	boolean	Build guidance based on the guidance document label	False	No
interactive_mode	Interactive Mode	boolean	Enable interactive learning, the assistant will build and apply guidance from updates and reprocess existing documents that aren’t guidance	False	No

---

​

Model Details

• Provider: Kodexa