Running Optimizations

Start an optimization to improve your prompts automatically.

Via Dashboard

1. Go to converra.ai/prompts
2. Select the prompt you want to optimize
3. Click Optimize
4. Configure options:
   • Mode: Exploratory (fast) or Validation (thorough)
   • Variants: How many alternatives to test (2-5)
   • Focus: What to improve (clarity, task completion, etc.)
5. Click Start Optimization

Via MCP (AI Assistant)

Simple:

Optimize my customer support prompt

With options:

Run an optimization on my support prompt with 3 variants,
focusing on improving clarity and task completion

With hypothesis:

Optimize my sales prompt. I think adding examples of objection
handling will improve conversion.

Via SDK

const optimization = await converra.optimizations.trigger({
  promptId: 'prompt_123',
  mode: 'exploratory',
  variantCount: 3,
  intent: {
    targetImprovements: ['clarity', 'task completion'],
    hypothesis: 'Adding step-by-step instructions will help'
  }
});

console.log('Started:', optimization.id);

Via API

curl -X POST https://converra.ai/api/v1/optimizations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "promptId": "prompt_123",
    "mode": "exploratory",
    "variantCount": 3
  }'

Configuration Options

Mode

Mode	Duration	Use Case
exploratory	5-15 min	Quick iteration, finding improvements
validation	30-60 min	Production decisions, statistical confidence

Variant Count

• 2 variants: Quick A/B test
• 3 variants: Balanced (recommended)
• 5 variants: Explore more options

Intent (Optional)

Guide the optimization:

{
  intent: {
    targetImprovements: ['clarity', 'conciseness', 'task completion'],
    hypothesis: 'Users struggle with technical terms',
    variationDegree: 'moderate' // 'minimal' | 'moderate' | 'significant'
  }
}

Custom Personas (Optional)

Test against specific user types:

{
  personaIds: ['persona_frustrated', 'persona_technical']
}

Monitoring Progress

Dashboard

Watch real-time progress at converra.ai/optimizations

Via MCP

How's my optimization going?

Show me the status of optimization opt_123

Via SDK

const status = await converra.optimizations.get('opt_123');
console.log(`Status: ${status.status}`);
console.log(`Progress: ${status.progress}%`);

What Happens During Optimization

1. Starting - Analyzing your prompt
2. Generating - Creating variant prompts
3. Simulating - Running test conversations
4. Evaluating - Measuring performance
5. Complete - Results ready

Stopping an Optimization

If needed, you can stop early:

await converra.optimizations.stop('opt_123', {
  reason: 'Need to modify the base prompt'
});

Tips for Better Results

1. Start with a decent prompt - Optimization improves, doesn't create from scratch
2. Be specific about goals - Use the intent parameter
3. Run exploratory first - Then validation for final decisions
4. Review variants - Learn from what works

Next Steps

• Understanding Results - Interpret your results
• Best Practices - Write optimization-ready prompts