The initial age of childhood represents a vital period for fostering nutritional well-being, and childcare settings provide a unique opportunity to influence dietary habits on a broad scale. This study introduces a Graph Attention Network (GAT) based approach to recommend top-N child-friendly recipes using the All_Recipes dataset. Recipes, ingredients, and user interactions are represented by nodes in a heterogeneous network, while connections such as ingredient similarity and user preferences are captured by edges. Suitability for young audiences is given priority in recommendations thanks to a child-friendliness score that is calculated using ratings, the number of raters, and review content. TF-IDF is used to process features including ingredients, preparation procedures, and nutritional content, which are then coupled with normalized numerical properties. By giving graph nodes attention weights, the GAT model may prioritize and discover important links while using feature data and graph structure to make better predictions. Our experiment’s findings show that the GAT-based model performs better in accuracy, recall, and NDCG metrics than baseline techniques like collaborative filtering and GCNs. This approach offers a scalable way to encourage young children to eat healthily by providing varied, pertinent, and customized recipe ideas.