Skip to main content
Learn how to generate AI outputs for your scenarios.

​
What Is Output Generation?

Output generation is when Sageloop calls your chosen AI model (GPT-4, Claude, etc.) to generate responses for each of your scenarios. How It Works:
  1. You add scenarios to your project
  2. You click “Generate Outputs”
  3. Sageloop calls your AI model once per scenario
  4. Results appear in a table for easy comparison

​
Getting Started

​
Prerequisites

  • Project Created: You’ve already created a project with a system prompt
  • Scenarios Added: You’ve added 10+ scenarios
Sageloop provides all the AI infrastructure you need. Just sign up and start generating.

​
Running Generation

  1. Navigate to your project
  2. Click “Generate Outputs”
  3. Select your model (GPT-4, GPT-3.5, Claude, etc.)
  4. Click “Generate”
  5. Wait for results (usually 30-60 seconds for 20 scenarios)
What You’ll See:
  • Progress indicator
  • Table of generated responses
  • Status for each scenario

​
Understanding Generation

​
Temperature & Consistency

Generation quality depends on your system prompt clarity:
  • Low temperature settings (0.0-0.5): Consistent, predictable responses
  • Higher temperature (0.7+): More diverse, creative variations
Use temperature 0.3-0.5 for evaluating support bots, content generation, etc. Use higher temperature (0.7+) if you want to test diverse variations.

​
Selective Regeneration

After your first generation, you can regenerate specific scenarios:
  1. Select scenarios you want to regenerate
  2. Click “Regenerate Selected”
  3. Only selected scenarios are regenerated (saves time)
When to Use:
  • After updating your system prompt
  • When you want to retry failed scenarios
  • When you want different variations of same scenario

​
Understanding Your Outputs

​
Output Table View

The output table shows:
  • Scenario (on left)
  • AI Response (in middle)
  • Your Rating (on right, if already rated)
Features:
  • Expand/collapse scenarios
  • Copy outputs to clipboard
  • Regenerate specific scenarios
  • Add notes/feedback

​
Looking for Patterns

Before rating, scan for patterns: Example:
  • Scenario 1: “Where is my refund?” → “Refunds take 5-7 business days”
  • Scenario 2: “Where is my refund?” → “Refunds arrive soon”
  • Scenario 3: “When will I get my refund?” → “Refunds are processed shortly”
Pattern: Inconsistent timeline wording (some say “5-7 days”, others say “soon”) Action: Rate accordingly and add feedback to help extraction identify this pattern.

​
Troubleshooting Generation

​
Problem: Generation Fails

Possible Causes:
  1. API rate limit exceeded
  2. Model not available
  3. Network connectivity issue
Fix:
  • Wait a few minutes and retry
  • Try a different model
  • Check your connection

​
Problem: Outputs Are Too Similar

Cause: Model configuration too rigid Fix: Regenerate with different settings or refine system prompt

​
Problem: Outputs Are Inconsistent

Cause: System prompt lacks specificity Fix: Add more detailed instructions to your system prompt, then regenerate

​
Problem: Outputs Are Too Short/Long

Solution: Update system prompt with output length guidance 
Your responses should be 150-250 words.
(instead of no guidance)
Then regenerate.

​
System Prompt Best Practices

​
Clear Instructions

✅ Good:
"You are a customer support agent. Always:
1. Start with an apology
2. Explain the policy clearly
3. Give specific timeline (5-7 business days)
4. Provide next steps
5. End with friendly offer to help"

❌ Vague:
"Be a helpful support agent"

​
Context

✅ Good:
"You are helping ShopCo customers with refund questions.
ShopCo's refund policy: 5-7 business days from receipt.
You cannot offer exceptions."

❌ Vague:
"Help with refunds"

​
Examples

✅ Good:
"When asked about refunds, respond like:
'I sincerely apologize for the inconvenience.
Refunds take 5-7 business days to process...'"

❌ Vague:
"Be empathetic"

​
Batch vs. Individual Testing

​
Sageloop (Batch)

Generate all 20 outputs at once
See them all in a table
Compare visually
Rate in bulk using shortcuts
Advantage: Fast pattern recognition

​
ChatGPT (Individual)

Test one output at a time
Hard to remember previous tests
Need to manually organize results
Disadvantage: Slow, patterns hidden

​
Next Steps