A newly launched platform by Belgian AI company PolyPerception enables recycling facilities to inquire about operational data using everyday language, with the support of Norwegian sorting specialist Tomra Recycling, which acquired a 51% stake in the company in May 2026.
PolyPerception, founded in Antwerp, Belgium in 2021, designed the AI-native platform specifically for material recovery facilities handling plastics, aiming to accelerate the identification of quality issues, material losses, and operational changes. The company’s technology utilizes cameras to monitor materials throughout the sorting process.
The platform introduces an AI-driven layer to the existing monitoring system, allowing operators to pose questions about plant performance and receive instant, data-driven responses, as explained by PolyPerception co-founder Parshva Mehta. For instance, an operator can inquire about the impact of a recent adjustment on recovery or purity rates, and the platform will analyze the relevant data to provide a clear answer.
Read Also: McCloskey unveils new waste shredder model
In one example, a material recovery facility monitoring polypropylene and polyethylene streams noticed an increase in bottles accumulating in the tray stream towards the end of the week. The platform detected a steady rise in contamination and flagged the issue within 45 minutes of its onset, demonstrating its ability to identify subtle changes that may elude manual detection in high-volume facilities.
The system can also generate quality reports, configure alerts tailored to specific plant requirements, and provide operators with actionable insights, according to Mehta. By doing so, the platform helps operators address problems before they affect output quality or material recovery.
Enhanced Search Capabilities
PolyPerception has incorporated search functions to accommodate evolving waste streams, including similarity searches for visually similar objects. Operators can identify a specific object, such as a vape, and search for similar items within the monitored waste stream, enabling the system to alert them when high-risk objects appear.
Read Also: London Homeowners Turn to Professional Clearance Services
Users can also search for specific objects or brands, such as refuse bags, nappies, or particular manufacturers, to monitor and address potential fire risks associated with batteries and electronic cigarettes. Mehta emphasizes that the system continuously collects production data, updating its deep-learning models every two weeks to ensure optimal performance.
PolyPerception adopts a gradual adoption approach, comparing its data with customers’ existing monitoring methods and highlighting areas where value may be lost in residual streams. According to Mehta, customers can start seeing returns on their investment within the first month of implementation.
The new AI solution and Tomra’s GainNext deep learning system are early steps in a larger transformation, according to Lars Enge, executive vice president at Tomra Recycling. Enge notes that deep learning solves previously impossible sorting challenges, such as identifying food-grade plastics, enabling recyclers to produce high-quality recyclates for contact-sensitive applications. He envisions that by 2036, materials will be tracked from input to output, providing greater transparency, traceability, and confidence in secondary raw materials. Machines, sensors, and systems will securely exchange information, creating more connected and intelligent recycling operations. As a result, recyclers will make decisions based on continuous data, removing operational uncertainty and making recycling more predictable and investment-friendly.