prompt-optimizer

You can optimize text classification prompts by running a seed prompt against labeled sample texts to initiate iterative, phase-driven improvements. The workflow refines label definitions anchored in real text patterns to reduce misclassification and generate production-grade prompts.

Evaluating LLM prompt performance involves using dev/test splits, macro-F1 scores, per-class metrics, and diagnostic confusion matrices. This phase-driven evaluation framework guides targeted improvements and helps identify specific classification errors across dimensions.

Yes, this prompt optimization workflow supports multi-dimension text classification. It provides per-dimension optimization and phase-driven diagnostics, allowing you to systematically refine prompts across multiple predefined categories simultaneously.

To reduce misclassification errors, the workflow uses immersion-based label grounding with error-driven refinement and focused re-immersion. This approach anchors label definitions in real text patterns and iteratively improves them based on diagnostic feedback.

The workflow generates deployment-ready batch processing templates in Python and R. These outputs include the final optimized prompt text, performance metrics, and structured documentation like prompt cards and method narratives for immediate integration.