iFood uses recommendation systems and machine learning models to improve the food delivery experience for customers. They generate word embeddings from text data to understand relationships between food items. A knowledge graph connects entities like users, restaurants, dishes and ingredients to enrich data. Graph convolutional networks are used to create embeddings that represent nodes based on their connections in the graph. These embeddings are used in recommendation systems at iFood to provide personalized search results, lists of similar items, and improve other machine learning models.