Testing & Fine-tuning Templates
Learn how to test and optimize your templates for better performance and accuracy.

# Step 1: Prepare Test Inputs
For the Geo Guesser example, you might test with inputs like:
- Famous city nicknames ("The Big Apple")
- Descriptive phrases ("It's a place that never sleeps")
- Landmark references ("the city holding the status of liberty")
- Images of landmarks (when using image input)
# Step 2: Run Tests in the Playground
Use the template playground to process each input and verify the results.
the city holding the status of liberty
{
"city": "New York",
"country": "USA",
"currency": "USD",
"isCapital": false,
"population": "8.4 million"
}
# Step 4: Analyze Results
When reviewing test results, check for these key aspects:
- Schema compliance: Does the output match your defined structure?
- Data accuracy: Is the extracted information correct?
- Completeness: Is all relevant information captured?
- Edge case handling: How does it handle unusual inputs?
- Processing time: How quickly does it complete?
# Fine-tuning Your Template
If your tests reveal issues or opportunities for improvement, you can fine-tune your template using the "Tune" tab.
# 1. Add or Refine Custom Prompts
Custom prompts can significantly improve template performance by giving the AI more specific instructions.

- "For population values, include the approximate number with 'million' or 'thousand'"
- "When uncertain about exact population, provide the most recent estimate"
- "If it's a fictional city like Gotham, set currency to 'bat$'"
- "For capital cities, always set isCapital to true"
# 2. Select the Optimal AI Model
Different models have different strengths. Try alternative models to see which performs best for your use case.

Model selection considerations:
- For complex tasks requiring nuanced understanding, try GPT-4o
- For faster response times and lower costs, consider GPT-3.5 Turbo
- For image analysis tasks, ensure you select a model with vision capabilities
# 3. Refine Your Schema
Sometimes issues stem from ambiguities in your output schema. Consider these improvements:

- Make field names clear and descriptive
- Choose appropriate data types for each field
- Use boolean fields (like isCapital) for yes/no data
- Consider adding more specific fields if needed
# Testing Workflow Best Practices
To ensure your template performs reliably in production:
- Test with a variety of inputs that represent real-world usage
- Include edge cases and potential ambiguities
- Test performance with different AI models
- Save successful test cases for regression testing
- Regularly re-test your template as you make changes
Once your template is performing well, you'll want to deploy it to make it available for use in your applications. Continue to Deployment.