LLM transform - Precisely Data Integrity Suite

Data Integrity Suite
Product
Spatial_Analytics
Data_Integration
Data_Enrichment
Data_Governance
Precisely_Data_Integrity_Suite
geo_addressing_1
Data_Observability
Data_Quality
dis_core_foundation
Services
Spatial Analytics
Data Integration
Data Enrichment
Data Governance
Geo Addressing
Data Observability
Data Quality
Core Foundation
ft:title
Data Integrity Suite
ft:locale
en-US
PublicationType
pt_product_guide
copyrightfirst
2000
copyrightlast
2026

The LLM Transform step enables you to utilize AI-powered transformations on your pipeline data with the help of Large Language Models (LLMs). This step can enhance, summarize, categorize, translate, or extract structured data from text fields.

Note: To run a pipeline on Snowflake that includes an LLM Transform step, you must define a Snowflake UDF. To define one, follow the instructions given here: Snowflake UDF for LLM Execution.
Limited Availability: This feature is currently available only in select workspaces and might be subject to change before general availability.

How to Add an LLM Transform Step

  1. Within your pipeline, click on the Add Step button.
  2. From the list of steps available, choose LLM Transform.
  3. The step will be added to the pipeline along with its configuration options.

How to Configure the LLM Transform Step

  1. Large Language Model
    • Select the LLM connection you wish to use.
    • If your administrator has configured a default model, it will appear automatically.
    • To change the model, open the Large Language Model dropdown and select from the available connections.
      Note: The availability of models depends on the connections defined in AI Manager.
  2. Input Fields
    • Select one or more fields from the source data for the model to transform.
    • Only the selected fields will be passed to the LLM.
  3. Transformation Instructions
    • Provide a prompt that describes how the model should transform the input.
    • Prompts can include multiple sentences or instructions.
    • The input box can expand up to a maximum height and then provides a scrolling option.
  4. Output Fields
    • Specify the field name(s) that will hold the transformed results.
    • You have the option to create a new field or overwrite an existing one.

    Example Use Cases

    Use Case Input(s) Prompt Output(s)
    Translation Non-English text field Translate to English New [Translation] field
    Sentiment Analysis Customer feedback or review field Identify the sentiment of the text: Positive, Negative, or Neutral New [Sentiment] field
    Classification Product description field Classify each item as ‘Electronics’, ‘Apparel’, or ‘Other’ New [Product Category] field

  5. Preview Results
    • Once you've configured the settings, click on Preview.
    • The preview panel will display input records alongside the generated output.
    • Review the results before saving the step.

Error Handling

  • Invalid Input Fields: The step requires at least one input field.
  • Unavailable Model: If the selected LLM connection is disabled, an error will be displayed. Please contact your administrator.
  • Prompt Issues: If the model cannot interpret the instructions, you may need to revise the prompt.

Best Practices

  • Use descriptive names for output fields to maintain clarity in results.
  • Always validate with Preview before saving.
  • Ensure prompts are short, specific, and unambiguous.
  • Test transformations on sample data before running them on a larger scale.