Reviewer Rules
Evaluate AI output quality with custom criteria
Reviewer Rules is a custom rule feature for evaluating and verifying AI output quality in emma's multi-agent workflow. Define your own evaluation criteria and achieve consistent quality management.
What are Reviewer Rules?
With Reviewer Rules, you can set your own criteria for evaluating AI output. Override standard evaluation aspects (completeness, accuracy, practicality) with custom rules to perform quality management that meets your business requirements.
Custom Evaluation Criteria
Set evaluation criteria based on industry-specific requirements or internal guidelines
Consistent Quality Management
Apply the same standards to evaluate all AI outputs
Use Cases
Compliance Check
Automatically verify compliance with legal requirements and industry regulations. Particularly effective in regulated industries such as finance, healthcare, and legal.
Brand Guideline Compliance
Evaluate whether outputs align with corporate tone & manner, glossary, and style guides. Ideal for marketing content quality management.
Technical Documentation Quality
Verify that technical documents such as API documentation and user manuals meet defined format and accuracy standards.
How to Create Rules
Access the Rules Page
Create a new rule from the Rules page in the emma dashboard.
Define Rule Content
Write the rule name, summary, and detailed instructions in Markdown format.
Apply in Playground
Select the created rule in Playground and apply it during task execution.
Rule Components
Rule Name
A name to identify the rule. Keep it clear and concise.
Summary
A brief sentence explaining the purpose of the rule.
Instructions
Detailed evaluation criteria in Markdown format. Include specific judgment conditions and checkpoints.
Version Control
Reviewer Rules has built-in version control functionality to track rule change history.
- -Automatically assign version numbers when rules are updated
- -View differences from past versions in a GitHub-style UI
- -Rollback to any version is possible