This research leverages Large Language Models and Retrieval-Augmented Generation techniques to automatically analyze unstructured visitation records, quantify child abuse risk levels, and provide precise intervention recommendations, significantly enhancing social workers＇ capability for timely intervention and optimal resource allocation.

This study pioneers globally by integrating seven internationally recognized child abuse risk assessment scales, employing advanced Large Language Models (LLM) and Retrieval-Augmented Generation (RAG) techniques to analyze visitation records, quantify abuse risks comprehensively, and provide culturally adapted, actionable recommendations, overcoming limitations of single-dimension international benchmarks.

This AI technology can be applied across social welfare, healthcare, education, and NGO sectors to automatically analyze visitation records, quantify abuse risks, and improve decision-making. It offers scalable deployment through a SaaS model with API integration into government systems. The framework has strong potential for international adoption and can be extended to related domains such as domestic violence and elder abuse prediction.