# LLM-Driven Evolutionary Search

Geospatial algorithm improvement has long relied on researchers hand-crafting each structural fix; an LLM-driven multi-agent framework now delegates candidate generation, evaluation and selection to automated search while the researcher defines the problem and the evaluation. Seeded with Two-stage K-Models and using the Rand Index against the ground-truth partition as fitness, the evolved algorithm was frozen and applied unchanged to 11 real datasets, reaching the lowest cross-dataset mean RMSE and the highest SSR Reduction across climate, health, hydrology and politics.

Geospatial algorithm improvement has long relied on researchers hand-crafting each structural fix; an LLM-driven multi-agent framework now delegates candidate generation, evaluation and selection to automated search while the researcher defines the problem and the evaluation. Seeded with Two-stage K-Models and using the Rand Index against the ground-truth partition as fitness, the evolved algorithm was frozen and applied unchanged to 11 real datasets, reaching the lowest cross-dataset mean RMSE and the highest SSR Reduction across climate, health, hydrology and politics.

The authors propose an evolutionary framework in which an LLM iteratively generates and refines natural-language binary features (rubrics), an LLM scores each sample against them, and the resulting vectors train a transparent classifier such as logistic regression; across AG News, GoEmotions, and HELOC, evolved features beat single-shot rubrics by +2.9 pp on average and outperform zero-shot LLM classification on two of three tasks, with the zero-shot LLM reaching only 50.7% on HELOC (near chance with a strong single-class bias) while evolved features reach 69.0% with balanced, auditable predictions.

The authors propose an evolutionary framework in which an LLM iteratively generates and refines natural-language binary features (rubrics), an LLM scores each sample against them, and the resulting vectors train a transparent classifier such as logistic regression; across AG News, GoEmotions, and HELOC, evolved features beat single-shot rubrics by +2.9 pp on average and outperform zero-shot LLM classification on two of three tasks, with the zero-shot LLM reaching only 50.7% on HELOC (near chance with a strong single-class bias) while evolved features reach 69.0% with balanced, auditable predictions.