Utilizing Retrieval-Augmented Large Language Models for Pregnancy Nutrition Advice

The importance of nutrition during pregnancy cannot be overstated, as it profoundly impacts maternal and fetal health outcomes. Optimal fetal growth and development are contingent upon adequate nutrition throughout gestation, which in turn requires that expectant mothers possess a high level of nutritional literacy. This latter factor may serve as a valuable predictor of pregnancy outcomes. This paper seeks to leverage the capabilities of a retrieval-augmented large language model to provide personalized prenatal nutrition guidance. We employed Meta’s LLAMA 2 model and integrated an expert-curated dataset of nutrition information. Our evaluation, conducted using ChatGPT-based metrics, revealed that while the augmented model did not yield significant improvements in overall response quality, it could generate more thoughtful and specific responses easily comprehensible to users. We conclude by discussing the challenges encountered and lessons learned from our investigation.