Abstract: FR-PO1092
Multiagent Retrieval-Augmented Generation-Based Artificial Intelligence (AI) System for Personalized Meal Recommendations in Patients on Hemodialysis
Session Information
- Health Maintenance, Nutrition, and Metabolism
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Health Maintenance, Nutrition, and Metabolism
- 1500 Health Maintenance, Nutrition, and Metabolism
Authors
- Yueh, Sheng-Han, Renal Research Institute, New York, New York, United States
- Bergling, Karin, Renal Research Institute, New York, New York, United States
- Usvyat, Len A., Renal Research Institute, New York, New York, United States
- Kotanko, Peter, Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Zhang, Hanjie, Renal Research Institute, New York, New York, United States
Background
Dietary planning is essential for patients with dialysis. However, existing recipe platforms offer limited opportunity for personalization. Retrieval-Augmented Generation (RAG), an AI method which integrates information retrieval with text generation, was initially applied but proved insufficient, yielding less relevant results and only a single recipe per query. To address these limitations, this study combines RAG with a multi-agent framework, enabling more accurate and diverse recipe recommendations and personalized meal planning.
Methods
923 kidney-friendly recipes were collected through web scraping and organized in a database. The system employs four agents (Fig. 1). The modification agent rewrites user query to expand the pool of potentially relevant recipes. Based on the revised query, the system retrieves the top 25 related recipes from the database. Simultaneously, the task agent determines whether the user is requesting a single meal or a meal plan and guides the workflow accordingly. For single recipe requests, the validation agent evaluates each retrieved recipe to ensure it meets the user's specific dietary requests. If a meal plan is requested, the generation agent analyzes the retrieved recipes and composes a plan that aligns with the user's needs.
Results
Compared to standard RAG approaches, this system enhances the overall quality of the retrieved results. By leveraging multiple specialized agents, it delivers higher query- output alignment, making recipe recommendations (Fig.1 Layout 1) and meal planning (Fig.1 Layout 2) more effective for users with complex dietary needs.
Conclusion
The framework improves dietary recommendation by enhancing three key areas: (1) expanding the pool of qualified recipes during the retrieval phase, (2) providing users with alternative meal options, and (3) increasing query- output matching. Collectively, these advancements support the development of a robust and adaptable platform for personalized dietary planning.
Figure1. Workflow of the solution with example queries.
Funding
- Commercial Support – Renal Research Institute LLC, a wholly owned subsidiary of Fresenius Medical Care Holdings Inc.