AI-driven personalized nutrition: RAG-based digital health solution for obesity and type 2 diabetes

Summary

This study presents an AI-driven dietary recommendation system using retrieval-augmented generation (RAG) and the LLaMA3 language model to generate personalized smoothie recipes for obesity and type 2 diabetes management. The system integrates evidence-based dietary guidelines from multiple authoritative sources with sustainability criteria, prioritizing plant-based and locally sourced ingredients aligned with UN Sustainable Development Goals. Evaluation of 1,000 recipes demonstrated strong adherence to nutritional targets and environmental criteria, with explainable AI features designed to improve health literacy and dietary adherence across diverse users.

Regional applicability

The system integrates guidelines from multiple international sources (RIVM, EUFIC, USDA, American Diabetes Association) and does not specify UK-based evaluation. Applicability to United Kingdom practice would require integration of UK-specific dietary guidelines (e.g. Eatwell Guide, NHS guidance) and consideration of local seasonal produce and food supply chains.

Key measures

Recipe adherence to health guidelines (80.1%), compliance with sustainability criteria (92%), explainability of AI recommendations, user engagement metrics

Outcomes reported

The system generated personalized smoothie recipes with 80.1% adherence to health guidelines (calories, fibre, fats) and 92% compliance with sustainability criteria. A prototype web application was developed to deliver real-time personalized dietary recommendations for chronic disease management.

Topic tags