Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Closed-Loop Multi-Agent System Driven by LLMs for Meal-Level Personalized Nutrition Management

Muqing Xu

TL;DR

This work tackles meal-level personalized nutrition management by integrating image-based logging with a closed-loop, LLM-driven multi-agent system. The proposed architecture distributes tasks across four agents (Controller, Dialog, Vision, File) to perform meal nutrition analysis, dynamic plan adjustment, and personalized recommendations, all anchored to USDA DRIs and the SNAPMe dataset. Experimental results demonstrate competitive nutrient estimation, effective personalization, and efficient task planning, while highlighting challenges in micronutrient quantification from images and the need for larger-scale studies. The approach advances mobile, automated nutrition guidance and points toward knowledge-grounded reasoning and retrieval-enhanced methods to improve precision and reliability in real-world use. Key findings include: (1) a robust closed-loop workflow that updates a daily intake budget and informs the next meal; (2) measurable planning efficiency with Plan Optimality (PO) ≈ 0.75 and end-to-end latency ≈ 65 s; (3) demonstrated personalization capabilities that adapt menus to user preferences and constraints; and (4) identified open challenges in micronutrient estimation and large-scale validation, guiding future work toward broader deployment and knowledge-grounded nutrition reasoning.

Abstract

Personalized nutrition management aims to tailor dietary guidance to an individual's intake and phenotype, but most existing systems handle food logging, nutrient analysis and recommendation separately. We present a next-generation mobile nutrition assistant that combines image based meal logging with an LLM driven multi agent controller to provide meal level closed loop support. The system coordinates vision, dialogue and state management agents to estimate nutrients from photos and update a daily intake budget. It then adapts the next meal plan to user preferences and dietary constraints. Experiments with SNAPMe meal images and simulated users show competitive nutrient estimation, personalized menus and efficient task plans. These findings demonstrate the feasibility of multi agent LLM control for personalized nutrition and reveal open challenges in micronutrient estimation from images and in large scale real world studies.

A Closed-Loop Multi-Agent System Driven by LLMs for Meal-Level Personalized Nutrition Management

TL;DR

This work tackles meal-level personalized nutrition management by integrating image-based logging with a closed-loop, LLM-driven multi-agent system. The proposed architecture distributes tasks across four agents (Controller, Dialog, Vision, File) to perform meal nutrition analysis, dynamic plan adjustment, and personalized recommendations, all anchored to USDA DRIs and the SNAPMe dataset. Experimental results demonstrate competitive nutrient estimation, effective personalization, and efficient task planning, while highlighting challenges in micronutrient quantification from images and the need for larger-scale studies. The approach advances mobile, automated nutrition guidance and points toward knowledge-grounded reasoning and retrieval-enhanced methods to improve precision and reliability in real-world use. Key findings include: (1) a robust closed-loop workflow that updates a daily intake budget and informs the next meal; (2) measurable planning efficiency with Plan Optimality (PO) ≈ 0.75 and end-to-end latency ≈ 65 s; (3) demonstrated personalization capabilities that adapt menus to user preferences and constraints; and (4) identified open challenges in micronutrient estimation and large-scale validation, guiding future work toward broader deployment and knowledge-grounded nutrition reasoning.

Abstract

Personalized nutrition management aims to tailor dietary guidance to an individual's intake and phenotype, but most existing systems handle food logging, nutrient analysis and recommendation separately. We present a next-generation mobile nutrition assistant that combines image based meal logging with an LLM driven multi agent controller to provide meal level closed loop support. The system coordinates vision, dialogue and state management agents to estimate nutrients from photos and update a daily intake budget. It then adapts the next meal plan to user preferences and dietary constraints. Experiments with SNAPMe meal images and simulated users show competitive nutrient estimation, personalized menus and efficient task plans. These findings demonstrate the feasibility of multi agent LLM control for personalized nutrition and reveal open challenges in micronutrient estimation from images and in large scale real world studies.
Paper Structure (13 sections, 9 equations, 6 figures, 6 tables)

This paper contains 13 sections, 9 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Closed-loop System Architecture
  4. Overall System Workflow
  5. Closed-Loop Dynamic Adjustment Strategy
  6. MAS Architecture
  7. Experiments and Results Analysis
  8. Datasets
  9. Meal Image Analysis Capability
  10. Nutritional Analysis Capability
  11. Personalization Capabilities
  12. MAS Task Planning and Completion Capability
  13. Conclusion

Figures (6)

  • Figure 1: Overview of the system workflow.
  • Figure 2: Closed-Loop Dynamic Adjustment Strategy
  • Figure 3: Overview of the MAS Multimodal Main Workflow
  • Figure 4: Personalized Next-Meal Recommendations for British Cuisine Preferences
  • Figure 5: Personalized Next-Meal Recommendations for Chinese Cuisine Preferences
  • ...and 1 more figures