Agri-food Waste Upcycling
Real World Challenge
The agri-food sector generates vast amounts of unavoidable by-products — from fruit pomace and brewery spent grain to processing residues — much of which is underutilised or sent to low-value uses. This wastes valuable nutrients, contributes to greenhouse gas emissions, and undermines the sector’s push towards net zero.
At the same time, global demand for sustainable protein is rising sharply. Traditional protein sources such as meat and dairy have high environmental footprints, and even plant-based proteins require land, water, and fertiliser. Upcycling agri-food waste streams into high-value microbial protein offers a double win: diverting waste from disposal and providing a low-carbon, circular alternative to conventional protein production.
The challenge lies in doing this efficiently and at scale. Waste streams are highly variable in composition, seasonality, and availability. Converting them into microbial protein requires careful control of fermentation conditions, precise nutrient balancing, and robust quality assurance — all while keeping costs competitive with existing protein sources.
Our AI Solution
At the Food AI Lab, we are building AI-driven data infrastructures and intelligent sensor systems to optimise the upcycling of agri-food waste into sustainable microbial protein.
IoT Sensors + Machine Learning – We combine low-cost online sensors with advanced ML algorithms to monitor and control fermentation in real time, achieving predictive accuracies of 95–100% across multiple industrial processes.
Adaptive Process Control – AI models can adjust fermentation parameters dynamically to account for the variability of incoming waste streams, ensuring consistent protein yields and quality.
Semantic Data Integration & Synthetic Data – To overcome gaps in real-world waste composition datasets, we integrate heterogeneous data sources and generate synthetic training data, enabling robust model performance without costly, large-scale trials.
Multi-Objective Optimisation – We balance protein yield with environmental and economic goals, reducing resource use and emissions without sacrificing product quality.
Federated & Transfer Learning – These methods allow models trained on one facility’s data to be improved using insights from others — without sharing proprietary data — accelerating adoption across diverse production sites.
By merging circular economy principles with AI-enabled process intelligence, our approach turns variable, low-value agri-food residues into a reliable source of sustainable protein, helping to decarbonise food production and close the loop on waste.
Partners and Funding: CSIRO, BBSRC and Bezos Earth Fund