Melodi uses a primary feedbackType which is typically “Positive” or “Negative”. Beyond this, you can define custom feedback attributes to capture detailed evaluations tailored to your project’s needs. These attributes are configured in your project settings.

You can define various feedback attributes and their corresponding value sets, such as:

  • Attribute: Star Rating
    • Values: 1, 2, 3, 4, 5 (stars)
  • Attribute: Quality Score
    • Values: “Excellent”, “Good”, “Fair”, “Poor”
  • Attribute: User Sentiment
    • Values: “Happy”, “Neutral”, “Frustrated”
  • Attribute: Content Relevance
    • Values: “Relevant”, “Partially Relevant”, “Irrelevant”
  • Attribute: Task Completion
    • Values: “Goal Achieved”, “Goal Not Achieved”
  • Any other categorical or numerical attribute relevant to your analysis (e.g., Humor Level: “High”, Issue Type: “Hallucination”).

This flexibility with attributes allows you to capture nuanced evaluations beyond the binary feedbackType, tailoring the labeling process to your specific AI agent and goals.

When creating a feedback record, it’s required to include either a feedbackType or at least one custom attribute. Providing both, or multiple attributes, is also supported.

When feedback is collected via the API or React component, you will specify the feedbackType (e.g., “POSITIVE”) and any relevant custom attributes you have configured (e.g., {"Star Rating": 5, "Quality Score": "Excellent"}).