Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Food4All: A Multi-Agent Framework for Real-time Free Food Discovery with Integrated Nutritional Metadata

Zhengqing Yuan, Yiyang Li, Weixiang Sun, Zheyuan Zhang, Kaiwen Shi, Keerthiram Murugesan, Yanfang Ye

TL;DR

Food4All tackles the real-world problem of access to free food for food-insecure populations by designing a dual-agent framework that unifies cross-source data aggregation, tool-grounded reasoning, and adaptive reinforcement learning. The Planner $\mathcal{A}_P$ and Executor $\mathcal{A}_E$ operate over a tool registry $\mathcal{F}$ in a five-module pipeline, enabling real-time queries to yield nutritionally annotated food options. An offline RL stage using Direct Preference Optimization with multi-component rewards improves geographic relevance, item accuracy, and nutrition fidelity, while online RL leverages user feedback to continually refine policy performance. Across large-scale evaluations and ablation studies, Food4All significantly outperforms baselines in end-to-end task success, retrieval accuracy, and trustworthiness, demonstrating practical impact for scalable, equitable food access and nutrition-aware guidance at the point of need.

Abstract

Food insecurity remains a persistent public health emergency in the United States, tightly interwoven with chronic disease, mental illness, and opioid misuse. Yet despite the existence of thousands of food banks and pantries, access remains fragmented: 1) current retrieval systems depend on static directories or generic search engines, which provide incomplete and geographically irrelevant results; 2) LLM-based chatbots offer only vague nutritional suggestions and fail to adapt to real-world constraints such as time, mobility, and transportation; and 3) existing food recommendation systems optimize for culinary diversity but overlook survival-critical needs of food-insecure populations, including immediate proximity, verified availability, and contextual barriers. These limitations risk leaving the most vulnerable individuals, those experiencing homelessness, addiction, or digital illiteracy, unable to access urgently needed resources. To address this, we introduce Food4All, the first multi-agent framework explicitly designed for real-time, context-aware free food retrieval. Food4All unifies three innovations: 1) heterogeneous data aggregation across official databases, community platforms, and social media to provide a continuously updated pool of food resources; 2) a lightweight reinforcement learning algorithm trained on curated cases to optimize for both geographic accessibility and nutritional correctness; and 3) an online feedback loop that dynamically adapts retrieval policies to evolving user needs. By bridging information acquisition, semantic analysis, and decision support, Food4All delivers nutritionally annotated and guidance at the point of need. This framework establishes an urgent step toward scalable, equitable, and intelligent systems that directly support populations facing food insecurity and its compounding health risks.

Food4All: A Multi-Agent Framework for Real-time Free Food Discovery with Integrated Nutritional Metadata

TL;DR

Food4All tackles the real-world problem of access to free food for food-insecure populations by designing a dual-agent framework that unifies cross-source data aggregation, tool-grounded reasoning, and adaptive reinforcement learning. The Planner and Executor operate over a tool registry in a five-module pipeline, enabling real-time queries to yield nutritionally annotated food options. An offline RL stage using Direct Preference Optimization with multi-component rewards improves geographic relevance, item accuracy, and nutrition fidelity, while online RL leverages user feedback to continually refine policy performance. Across large-scale evaluations and ablation studies, Food4All significantly outperforms baselines in end-to-end task success, retrieval accuracy, and trustworthiness, demonstrating practical impact for scalable, equitable food access and nutrition-aware guidance at the point of need.

Abstract

Food insecurity remains a persistent public health emergency in the United States, tightly interwoven with chronic disease, mental illness, and opioid misuse. Yet despite the existence of thousands of food banks and pantries, access remains fragmented: 1) current retrieval systems depend on static directories or generic search engines, which provide incomplete and geographically irrelevant results; 2) LLM-based chatbots offer only vague nutritional suggestions and fail to adapt to real-world constraints such as time, mobility, and transportation; and 3) existing food recommendation systems optimize for culinary diversity but overlook survival-critical needs of food-insecure populations, including immediate proximity, verified availability, and contextual barriers. These limitations risk leaving the most vulnerable individuals, those experiencing homelessness, addiction, or digital illiteracy, unable to access urgently needed resources. To address this, we introduce Food4All, the first multi-agent framework explicitly designed for real-time, context-aware free food retrieval. Food4All unifies three innovations: 1) heterogeneous data aggregation across official databases, community platforms, and social media to provide a continuously updated pool of food resources; 2) a lightweight reinforcement learning algorithm trained on curated cases to optimize for both geographic accessibility and nutritional correctness; and 3) an online feedback loop that dynamically adapts retrieval policies to evolving user needs. By bridging information acquisition, semantic analysis, and decision support, Food4All delivers nutritionally annotated and guidance at the point of need. This framework establishes an urgent step toward scalable, equitable, and intelligent systems that directly support populations facing food insecurity and its compounding health risks.
Paper Structure (20 sections, 18 equations, 4 figures, 4 tables, 2 algorithms)

This paper contains 20 sections, 18 equations, 4 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Food Insecurity
  4. Multi-Agent Information Retrieval System
  5. Methodology
  6. Multi-Agent Framework
  7. Offline Reinforcement Learning
  8. Online Reinforcement Learning
  9. Evaluation Principles
  10. Experiment
  11. Setting
  12. Offline Reinforcement Learning Evaluation
  13. Online Reinforcement Learning Evaluation
  14. Ablation Study
  15. Case Study
  16. ...and 5 more sections

Figures (4)

  • Figure 1: Overview of food insecurity and limitations of existing assistance systems. (a) Food insecurity stems from socioeconomic inequities and leads to adverse health outcomes. (b) Current chatbot and search-agent systems provide incomplete, outdated, or non-personalized information, underscoring the need for adaptive, context-aware solutions.
  • Figure 2: Overview of the Food4All framework. A dual-agent system where the Planner decomposes queries and the Executor performs tool-based actions. Validated results are stored in conversational memory until completion, producing structured, nutrition-aware food recommendations.
  • Figure 3: Radar chart illustrating Food4All performance across eight evaluation metrics under different online reinforcement learning (ORL) update steps. All metrics were normalized to the [0.8, 1.0] interval (with MiniDis inversely scaled) to highlight relative improvements over the baseline.
  • Figure 4: Comparison across key components using (left) subtask-level metrics, (middle) radar chart, and (right) qualitative heatmap visualization. Radar chart and heatmap metrics are normalized to the [0.5, 1] range for consistent visualization (with MiniDis inverted for interpretability). Results highlight that removing the Executor Tools or Planner Agent modules causes the most significant degradation across overall performance dimensions.