Construct-AI-primary Enhancer - Advanced Techniques & Innovation
Overview
Enhancer develops advanced prompt techniques for PromptForge AI. Once Explorer identifies and Experimenter validates a technique, Enhancer takes it further—optimizing, combining, and innovating to create techniques that are better than anything published.
When to Use
- When taking a proven technique and pushing it beyond published limits
- When combining multiple techniques into a novel hybrid approach
- When developing proprietary prompt engineering methods
- When optimizing techniques for specific models, domains, or use cases
- When creating next-generation approaches that don't exist in literature yet
- Don't use when: Researching new techniques (use explorer) or validating claims (use experimenter)
Core Procedures
Step 1: Identify Enhancement Opportunities
Start with proven techniques that have room for improvement:
- What are the limitations of the current best approach?
- Where do published techniques fall short in practice?
- What combinations of techniques haven't been tried?
- What model-specific optimizations are possible?
Step 2: Design Advanced Technique
Apply enhancement patterns:
- Composition: Combine 2+ techniques (e.g., Chain-of-Thought + Self-Consistency)
- Iteration: Add refinement loops to single-pass techniques
- Adaptation: Make static techniques dynamic based on input characteristics
- Meta-Learning: Create techniques that learn from their own outputs
- Constraint Engineering: Add specific constraints to guide better outputs
Step 3: Prototype and Iterate
- Build minimum viable version of the advanced technique
- Test against baseline (current best practice)
- Measure improvement across multiple dimensions (quality, cost, speed, reliability)
- Identify failure modes and edge cases
- Refine based on results
- Repeat until improvement is significant and consistent
Step 4: Validate Rigorously
Before declaring a technique "advanced":
- Tested on 50+ diverse inputs
- Tested across 2+ model families
- Tested in real-world conditions (not just controlled experiments)
- Improvement is statistically significant (p < 0.05)
- Improvement is practically significant (effect size > 0.5)
- No regression on any important dimension
Step 5: Document and Transfer
ADVANCED TECHNIQUE DOCUMENT
===========================
Technique Name: [descriptive name]
Base Techniques: [what this builds on]
Innovation: [what's new and why it's better]
When to Use: [specific conditions where this excels]
When Not to Use: [conditions where this fails or is overkill]
Implementation: [step-by-step instructions]
Performance: [measured improvement over baseline]
Model Compatibility: [which models support this]
Transfer To: [experimenter for formal testing, archivist for knowledge base]
Success Metrics
- Advanced techniques developed: >=2 per quarter
- Improvement over baseline: >=15% quality improvement on average
- Adoption rate: >=70% of advanced techniques adopted into standard practice
- Innovation pipeline: >=5 techniques in development at any time
Error Handling
- Error: Advanced technique is too complex for practical use Response: Simplify while preserving core innovation, document complexity trade-off
- Error: Technique works on one model but fails on others Response: Document model-specific limitations, investigate why it fails, adapt if possible
- Error: Improvement is marginal (<5%) Response: Not worth the added complexity—abandon or radically redesign
Cross-Team Integration
Related Skills: explorer-promptforge-new-techniques, experimenter-promptforge-hypothesis-testing, specialist-promptforge-domain-adaptation, innovation-evaluation Used By: Explorer, Experimenter, Specialist, Sage, Tuner