Field Container Instance - AI Correct

Performs AI-powered correction of field data using user-provided instructions.

Remarks

The AI Correct command enables automated correction of extracted field data within a Field Container Instance by leveraging large language models (LLMs). It is designed for scenarios where traditional extraction or validation methods are insufficient, allowing users to specify custom instructions that guide the AI in correcting claim or field values according to business rules, formatting requirements, or other domain-specific guidelines.

Role and Usage

The AI Correct command is used to apply AI-driven corrections to extracted data in fields, sections, or tables within Grooper. It is particularly useful when data quality requirements exceed the capabilities of standard extraction or validation logic, or when business rules are too complex or variable to encode directly. By configuring the command with a Data Generator and providing detailed instructions, users can direct the AI to interpret, validate, and correct field values according to their specific needs.

Typical use cases include:

• Normalizing data formats (e.g., dates, currency, identifiers)
• Applying business logic or conditional corrections
• Filling in missing or ambiguous values based on context
• Enforcing consistency and compliance with organizational standards

Configuration Guidance

To use AI Correct effectively:

• Set the 'Generator' property to a Data Generator configured for your preferred LLM provider and model.
• Write clear, specific instructions in the 'Instructions' property to describe the desired correction logic.
• Enable 'ApplyToAll' to process all claims in the collection, or disable it to correct only the current claim.
• Use 'IncludeContent' and 'QuotingMethod' to provide relevant document content as context for the AI, which can improve correction accuracy.

The quality and specificity of the instructions directly impact the results. Review and refine instructions as needed to achieve the desired correction behavior.

How It Works

1. The command gathers the current field data and schema from the Field Container Instance.
2. It constructs a prompt that includes user-provided instructions and, optionally, quoted document content.
3. The prompt is submitted to the configured Data Generator, which interacts with the LLM.
4. The LLM returns a set of JSON patch operations describing the required corrections.
5. The command applies these patches to the field data and performs validation to ensure data integrity.

Diagnostic Artifacts

When executed, AI Correct logs diagnostic information, including:

• The JSON schema used for the claim data (Schema.json)
• The number of correction operations performed
• A log of the LLM chat interaction, including the prompt sent and the response received
• Any errors or validation issues encountered during correction

These artifacts can be reviewed to troubleshoot issues or audit the correction process.

Considerations

• The effectiveness of AI-powered correction depends on the quality of the LLM, the clarity of instructions, and the relevance of included document content.
• Ensure that the selected Data Generator is properly configured and accessible.
• Corrections are validated after application, but users should review results to confirm accuracy and compliance with business requirements.

Properties

- Normalize all dates to MM/DD/YYYY format.
- Ensure all amounts are positive numbers with two decimal places.
- If the "Policy Number" is missing, set it to "UNKNOWN".
- Add a 2% adjustment to each claim.

See Also

Data GeneratorDI LayoutJSON FileMulti QuoteLabeled RegionLayout ObjectsData ValuesExtractedSemanticSection Instance CollectionSection InstanceTable Row InstanceTable InstanceDocument Instance