Introduction: Obesity in children and adolescents has been a global concern. In recent years, artificial intelligence has advanced in the medical field and achieved significant outcomes. Objectives: This study aims at developing a large-language model for aiding management of obesity in childhood and adolescents, Fuel Star, enabling intelligently interactive and comprehensive management through multi-scenario - ""hospitalschool-home"" - in-depth consultation. Methods: A multi-dimensionally structured medical dataset for obesity was constructed, via prompt engineering to cover typical tasks. The Retrieval-Augmented Generation (RAG) technical framework was introduced to enhance semantic consistency . Results: A total of 61 guidelines, 87 expert consensuses, 1,911 research articles, and 205 highquality clinical cases were included. Through document parsing modules that identified and extracted structural elements, data from different sources were standardized in format and aligned semantically . Each semantic unit was encoded into multi-dimensional vector representations by an embedding model and stored in a semantic-vector database. Prompt engineering guided the model to effectively invoke domain knowledge and control responses without modifying internal parameters, covering typical task types. The RAG framework enabled high-quality semantic retrieval at the document level and, through similarity reevaluation and priority adjustment, ensured maximum alignment between output content and user intent, thereby generating responses that integrated contextual and knowledge-based information. Conclusion: A multi-source professional database built in Fuel Star, combined with high-quality semantic modeling and lightweight adaptation strategies, supports multi-scenario application, rapid decision-making, in-depth consultation, and interactive question-answer, contributing to the comprehensive long-term management of obesity in children and adolescents.