Digital pantry: probabilistic machine learning for predicting onion shelf life

Overview

Digital Pantry is a data-driven predictive modeling solution that maps onion grower field data and growing conditions to a probabilistic model of shelf life over time. Developed by digiLab, the approach uses advanced machine learning to deliver predictions alongside quantified uncertainty, making it suitable for real-world agricultural data that is often noisy and high dimensional. The solution is positioned as an intelligent pantry, helping fresh produce stakeholders anticipate how long onions will remain viable in storage, reduce waste, and optimize supply chain decisions.

The offering is built on digiLab's Uncertainty Engine, a cloud-based AI platform designed for safety-critical and regulated industries. By combining deep Gaussian process modeling with explainable workflows, Digital Pantry brings enterprise-grade predictive analytics to the fresh produce sector, where data quality and variability have historically limited the usefulness of conventional AI.

Technical specifications

Core technology:

• Probabilistic machine learning based on deep Gaussian processes
• Uncertainty quantification built into every prediction, enabling deployment in regulated and high-stakes environments
• Robustness to noisy, sparse, and incomplete real-world data

Methodology:

• Automated cleaning of noisy input data from grower records and growing conditions
• Feature learning to identify the inputs that most strongly drive shelf life predictions
• Verification, Validation and Uncertainty Quantification (VVUQ) of the resulting model

Deployment:

• Cloud-based delivery through digiLab's Uncertainty Engine platform for fast, secure rollout
• Designed for usability by non-expert users in future iterations
• Workflow informed by digiLab's track record in water, aviation, and fusion energy sectors

Benefits for partners:

• Data-efficient modeling that performs well even where grower datasets are limited
• Transparent, explainable predictions suitable for regulated supply chain applications
• Reduced produce waste through better shelf life forecasting
• Scalable cloud infrastructure with no requirement for in-house AI expertise

Technology readiness level

Digital Pantry is at an early-to-mid stage of development. The underlying probabilistic machine learning methods and the Uncertainty Engine platform are validated through prior deployments in regulated industries including fusion energy (UKAEA plasma turbulence modeling), aviation, and water sectors. The onion shelf life application itself is currently at the research and feasibility stage, with a proposed validation programme estimated at approximately three months, contingent on data quality. The next steps involve consultation, data gathering, and the three-stage delivery workflow of automated cleaning, feature learning, and VVUQ before the model can be deployed commercially.