This paper proposes an artificial intelligence model to support the development of a personalized nutrition ecosystem for preschool children. The proposed approach combines child clustering based on physical characteristics, multi-objective meal plan optimization, and machine learning models to support decision-making. The system is designed as an open ecosystem that enables data collection from schools and families to facilitate continuous learning and model updates. Experiments were conducted using data from 150 children over a 10-week period in Nghe An Province, Vietnam. The results indicate that the proposed approach improves the nutritional adequacy of meal plans, increases food consumption rates, and reduces food waste compared with the current approach.