LLM Agent for Automated Feature Extractor Synthesis

This agentic framework uses an LLM-driven check-fix-verify loop to synthesize executable Python scripts for constraint satisfaction problem feature extraction. The method automates the creation of interpretable, problem-specific structural properties like graph density and constraint tightness from MiniZinc models.