AI-Powered Self-Improvement

Automatic Prompt
Optimization

Watch AI improve its own lead-ranking instructions. The system analyzes mistakes, learns patterns, and iteratively refines the prompt for better accuracy.

Evaluate

Run current prompt on 50 test leads

Analyze Errors

Find false positives & negatives

Generate Fixes

AI suggests prompt improvements

Apply & Repeat

Update prompt, iterate until optimal

Optimization Status

Ready to Optimize

Run a 5-iteration improvement cycle. The system will benchmark against the Evaluation Set and evolve instructions from v1 to v5.

Takes 5-10 minutes — no need to wait around

Do nothing to use the pre-loaded default dataset.
Expected columns: Full Name, Title, Company, Employee Range, Rank

Optimization Strategy

Objective

Maximizing F1-Score (Relevance) and NDCG@3 (Ranking Accuracy) on a held-out evaluation set of 50+ labelled leads.

Methodology

Using Textual Gradient Descent. The system identifies specific cases where the prompt failed, generates a "critique", and rewrites instructions to fix edge cases without breaking existing behavior.

Current Active Prompt

This is the live system prompt currently being used to score leads. The optimization engine iteratively improves this text based on evaluation results.

system_prompt_v1.mb

// No active prompt found // Initialize system to generate first version...