Ai-powered predictive design for accelerated food formulation

Overview

The AI-powered predictive design for food formulation revolutionizes the way recipes are developed by significantly reducing the number of experimental iterations required. By utilizing bespoke machine learning models, this solution connects ingredient properties and processing parameters with key product performance attributes such as taste, texture, and nutrition. With the integration of AI, businesses can achieve smarter, data-driven recipe designs from the concept stage, optimizing formulations and minimizing costs associated with bench testing.

Technical specifications

The technology employs bespoke machine learning models, including quantitative structure-performance relationship (QSPR) models and potentially generative AI, to predict formulation performance. The models are trained on historical formulation data to automate model building and validation, inspired by AutoML principles. This allows for rapid in silico predictions, optimization under constraints, ingredient substitution simulations, and generation of promising starting recipes. Key features include:

• Automated model building and validation using Python workflows
• In silico prediction of formulation performance
• Simulation of ingredient substitutions and generation of high-potential recipes
• Optimized experimental design for minimal bench testing

Technology readiness level

The technology is currently at TRL 5, having been validated through practical deployment in industry settings. The development plan includes three phases: foundation and scoping, iterative model refinement, and optimization and deployment, with ongoing improvements through active learning and high-value bench testing suggestions.