Summary
Risk assessments and other models provide a unique opportunity to improve food safety decision making in the produce industry. While several relevant models have been developed and published, use of these models in practical real-world decision making is still limited. In particular, it is challenging for industry to rapidly identify and access the appropriate models to support decision making in case of contamination events or possible deviations (e.g., temperature abuse) that may impact food safety. This project will address these issues and take advantage of (i) an existing standardized model repository and (ii) artificial intelligence (AI) tools to facilitate and expand industry use of produce safety modeling tools. Key outcome will include the availability of an AI agent (like a Chatbot, but substantially more sophisticated and designed for produce safety modeling applications) that will allow for conversational input (e.g., “I want to run a model to assess ....”) to run validated models. These modeling and model deployment efforts will be combined with a produce rapid response team that will investigate produce food safety issues to (i) identify root causes and (ii) collect new data and test our modeling infrastructure for its practical ability to provide relevant answers for risk management.
Technical Abstract
A substantial number of research studies have developed risk assessment models that describe pathogen transmission and intervention effectiveness across the produce supply chain for specific produce items (e.g., different leafy greens, cantaloupe) with most of them assessing public health risks (such as number of illnesses linked to a specific commodity). However, these models remain difficult to execute and interpret in real-world decision contexts due to (i) heterogeneous implementations, (ii) limited accessibility, and (iii) technical expertise required for scenario-specific use. Recent advances in agentic artificial intelligence (AI) and large language models (LLMs) enable natural-language interaction with external software systems, providing a pathway to make risk assessment models more accessible to industry and regulatory stakeholders. The research proposed here will leverage the existing Risk Assessment Knowledge Integration Platform (RAKIP) model repository to provide access to produce-focused microbial risk assessment models; as part of this process, models will be curated, standardized, and made directly executable through an agentic AI interface that we will develop. The agentic system will allow users to efficiently (i) discover relevant models, (ii) execute simulations, and (iii) interpret results using natural language, with responses grounded in produce-relevant scientific literature and model documentation. This system will be designed to ensure user confidentiality by not storing user inputs or using them for AI model training. We will also build a standardized scenario library that controls the execution of industry-relevant, modifiable scenarios within microbial risk assessment models. Finally, a rapid response team will be established, including defined procedures and trained roster, to investigate produce food safety incidents, identify root causes, collect new data, and refine models and scenarios. This will allow for real-world validation, testing, and improvement of the framework developed. Specific objectives include:
Obj. 1: Leverage the existing RAKIP model repository to curate produce-related microbial risk assessment models for predicting recall and public health risks.
Obj. 2: Develop an agentic AI interface that connects to the model repository to support discovery, execution, and interpretation of produce-related microbial risk assessment models through natural-language interactions.
Obj. 3: Curate a standardized scenario library of industry-relevant food safety “what-if” scenarios to support controlled and reproducible evaluation of realistic produce safety decision contexts.
Obj. 4: Develop rapid response team procedures and roster and deploy teams to at least three contamination events to investigate root causes, collect model input data, and test and validate models.
Beyond delivering a decision-support tool, we aim to establish a novel framework for proactive food safety management, enabling stakeholders to target effective controls while reducing resources spent on low-impact practices.
Research Objectives
Obj. 1: Leverage the existing RAKIP model repository to curate produce-related microbial risk assessment models for predicting recall and public health risks.
Obj. 2: Develop an agentic AI interface that connects to the model repository to support discovery, execution, and interpretation of produce-related microbial risk assessment models through natural-language interactions.
Obj. 3: Curate a standardized scenario library of industry-relevant food safety “what-if” scenarios to support controlled and reproducible evaluation of realistic produce safety decision contexts.
Obj. 4: Develop rapid response team procedures and roster and deploy teams to at least three contamination events to investigate root causes, collect model input data, and test and validate models.
Findings & Recommendations
This project is ongoing. A final report will be provided when the project is finished.