Human donor milk provides critical nutrition for the millions of infants that are born preterm each year. While the macronutrients in donor milk are critical to infant development, they vary by donation. In collaboration with Rogers Hixon Ontario Human Milk Bank, we developed a data-driven framework to pool multiple donations using machine learning and optimization. Over a one-year trial, our implementation yielded significantly higher macronutrient content than current pooling practices, with the proportion of pools meeting clinical fat and protein targets increasing by approximately 31%, with a 60% decrease in recipe creation time.