Prompt Optimization

Build Prompt Families that automatically select the best prompt for each input. Create specialized prompts that work together to perform better than any single prompt could.

How Prompt Families Work

Instead of using one prompt for everything, Empromptu creates Prompt Families - collections of prompts that work together. Start with one prompt, and the system helps you build out the whole family.

❌Traditional Approach

Single Prompt → All Inputs → Inconsistent Results

✅Empromptu Approach

Input Analysis → Best Prompt from Family → Optimized Result

4 Optimization Interfaces

Access all interfaces through the Actions button on any task. Each tab handles different aspects of prompt optimization.

📊Event Log

Track every optimization attempt with detailed information:

Timestamp: When the optimization occurred

Input used: What text was processed

Model: Which AI model was used (e.g., gpt-4.1-mini)

Temperature: Model settings (e.g., 0.2)

Response: The generated output

Score: Performance rating (0-10)

Score Reasoning: Detailed explanation of why the score was assigned

👥Prompt Family

View and manage your collection of specialized prompts:

Family Overview: See all prompts working together

Individual Prompts: View each family member with creation date

Status Control: Activate/deactivate prompts

Performance Tracking: See which prompts perform best

🔧Manual Optimization

Step-by-step improvement wizard for targeted enhancements:

Step 1: Select prompts to evaluate and optimize

Step 2: Choose test inputs for scoring

Step 3: Select evaluation criteria to optimize

Results: Generate and test prompt variations

⚡Automatic Optimization

Let the system optimize for you with minimal input:

Analysis: System reviews current Prompt Family performance

Generation: Creates new prompt variations automatically

Testing: Runs new prompts against inputs and evaluations

Selection: Adds best-performing prompts to your family

Understanding Optimization Scores

0-10 Scoring Scale

9.0-10.0

Excellent

Optimal performance, ready for business use

7.0-8.9

Good

Business-ready, minor edge cases

5.0-6.9

Acceptable

Improvable but functional

3.0-4.9

Needs Attention

Moderate issues, requires optimization

0.0-2.9

Poor

Requires immediate optimization

Score Reasoning Examples

"extracted_completeness - AI response captures the essence and key emotional elements. Summary captures key experiential aspects while maintaining vivid language."

Score: 7.000

Each score includes detailed reasoning that explains which evaluations passed/failed, specific issues identified, and what contributed to the score.

Optimization Strategies

Start with Automatic

Best for establishing baseline performance and handling common optimization patterns.

1. Click "Start Automatic Optimization"

2. System analyzes current performance

3. New prompts generated and tested

4. Results appear in Event Log

5. Prompt Family grows automatically

Refine with Manual

Best for addressing specific weaknesses and fine-tuning performance.

1. Review Event Log to identify problem areas

2. Select specific inputs that performed poorly

3. Choose evaluation criteria to focus on

4. Run targeted optimization experiments

5. Add successful variations to family

Building Effective Prompt Families

Include Different Scenarios

•Different input types (positive/negative content)
•Different output requirements (brief/detailed)
•Different edge cases (unusual inputs)

Use Representative Test Data

•Test data should match real use cases
•Include edge cases and challenging scenarios
•Sufficient volume to identify patterns

Monitor Performance Continuously

•Check Event Log regularly for new patterns
•Update families as use cases evolve
•Remove underperforming prompts