​
Overview

Latitude’s prompt editor allows you to define the expected JSON structure for your prompt’s output directly in the frontmatter. This feature ensures that the AI model returns responses in a consistent, structured format that can be easily parsed and used in your applications.

​
Specifying JSON Output

To specify the JSON output structure, use the output property in the frontmatter of your prompt. This property defines the structure and types of the expected output.

​
Basic Syntax

Here’s the basic structure for specifying JSON output in the frontmatter:

provider: <provider_name>
model: <model_name>
schema:
    type: object
    properties:
        <property_name>:
            type: <property_type>
        required:
            <required_property_name>

Example Here’s an example of how you might specify a JSON output for a sentiment analysis prompt:

---
provider: openai
model: gpt-4
schema:
    type: object
    properties:
        sentiment:
            type: string
            enum: [positive, negative, neutral]
        confidence:
            type: number
            minimum: 0
            maximum: 1
        explanation:
            type: string
        required:
            sentiment
            confidence
---

In this example, we’re expecting the AI to return a JSON object with three properties:

  • sentiment: a string that must be either “positive”, “negative”, or “neutral”
  • confidence: a number between 0 and 1
  • explanation: a string (optional)

The sentiment and confidence fields are required in the output.

​
Benefits of Using JSON Output

  • Consistency: Ensures that the AI model always returns data in the expected format.
  • Validation: Latitude can automatically validate the AI’s response against the specified structure.
  • Ease of Integration: Structured JSON output is easier to integrate into your applications and workflows.
  • Clear Expectations: Helps guide the AI model to provide responses in the desired format.