Solving Food Waste with Edge AI

Mill addresses the global food waste crisis—a multi-hundred billion dollar problem accounting for 10% of global emissions—by combining hardware engineering with AI. The company’s approach moves beyond simple composting by using computer vision to identify, categorize, and quantify food waste in real-time. This data allows commercial kitchens to optimize procurement and menu planning, effectively preventing waste before it occurs.

Edge Architecture and Model Tuning

To handle the high-volume data required for accurate identification, Mill deploys custom-tuned Gemma models directly onto Nvidia Jetson devices integrated into their hardware. This edge-first architecture is a strategic choice to avoid the prohibitive cost and latency of streaming high-frame-rate video (120–240 FPS) to the cloud. By running inference locally, the system maintains privacy and operational efficiency. The team leverages a massive dataset—over 5 terabytes of labeled food waste imagery—to fine-tune these models, ensuring they can distinguish between complex, messy waste items that would be difficult for off-the-shelf models to identify.

From Data to Agentic Workflows

Beyond simple identification, Mill aims to automate the administrative burden on kitchen staff. By analyzing waste patterns, the system generates predictive insights that feed into agentic procurement workflows. For example, if the system detects an impending surplus of a specific ingredient, it can suggest menu adjustments to chefs. This transforms the trash bin from a passive receptacle into an active data source, helping businesses save money while reducing their environmental footprint. The company emphasizes that while high-complexity models are necessary for critical fields like medicine, their application prioritizes rapid, iterative learning where the cost of a minor identification error is low compared to the massive aggregate value of the insights generated.