The Center for Data Innovation recently spoke with Daniel Oldak, Chief Strategic Officer of Clean Plate Innovations, a Pittsburgh-based company using machine learning to reduce food waste in restaurants and institutional dining settings. Oldak explained how the company’s system identifies what diners leave behind, measures waste volume, and provides food-service operators with actionable data to cut costs and reduce their environmental footprint.
David Kertai: What problem is Clean Plate Innovations solving?
Daniel Oldak: In the United States, an estimated 30 to 40 percent of all food produced for grocery stores, restaurants, and cafeterias is never eaten and ultimately ends up in the trash each year. Institutional cafeterias—such as school lunchrooms, university dining halls, hospital cafeterias, and museum cafés—generate roughly five million tons of food waste annually. That waste costs the food-service industry about $11 billion every year and produces approximately 170 million tons of greenhouse gas emissions annually, the equivalent of Air Force One circling the globe 130,000 times.
This waste comes from two main sources: food discarded during kitchen preparation, known as kitchen waste, and food diners leave on their plates, known as plate waste. Clean Plate Innovations focuses specifically on plate waste.
Our goal is to track and analyze what diners throw away so cafeterias can understand which foods consistently go uneaten, whether portion sizes are too large, and what operational changes could reduce waste. To do this, we built a system that uses computer vision to automatically identify and measure leftover food. By turning plate waste into structured, real-time data, we give dining managers, sustainability teams, and procurement staff clear insight into where waste occurs and how they can meaningfully reduce it.
Kertai: How does your system recognize the food on a plate?
Oldak: We install our camera system directly above waste‑disposal areas, where diners discard their plates after eating. The system uses a RealSense 3D imaging camera array—depth‑sensing cameras designed to capture object shape and spatial layout—paired with an AI model trained on more than 11 million images of food. As plates pass beneath the camera, it captures both the visual appearance of leftover food and its three‑dimensional volume.
The AI model then identifies each food item and estimates how much of it remains by calculating its size and density. That information is immediately translated into structured data, recording not just what was wasted but how much of each item was discarded and how often that occurs across meals, menus, and days. This level of measurement is critical because accurate volume data, not just surface appearance, determines the true cost and impact of food waste.
For example, two plates can look similar from above: one might have a thin spread of rice while another holds a compact portion of meat. Visually they appear comparable, but the actual amount of food wasted, and its financial and environmental impact, can be very different. By measuring volume rather than relying on appearance alone, the system determines how much food diners truly discard. The 3D mapping also separates overlapping items, distinguishes food from non‑food materials, and stays accurate even when plates are tilted or partially obscured, allowing the system to process thousands of plates per hour and provide a scalable, near real‑time view of waste.
Kertai: How do you improve waste identification accuracy over time?
Oldak: Our system improves through a combination of training data, human feedback, and exposure to new foods. During initial deployment, we rely heavily on the large dataset used to train the model. When the system encounters an item it cannot confidently identify, it flags the image for review. Our team then labels the item and verifies its weight in controlled test conditions, and then feeds that information back into the model.
As we deploy the system in more locations, it encounters a wider range of foods, serving styles, and cultural cuisines. Data from one site helps improve performance at others, allowing the system to become more accurate and adaptable over time.
Kertai: What sets you apart from other food waste tracking tools?
Oldak: Many organizations still rely on manual waste audits, which require staff to move heavy bins of discarded food and sort items by hand. These audits are labor‑intensive, unpleasant, and usually conducted only occasionally, so they capture only a narrow snapshot of what’s actually happening. A single measurement can easily be misleading, for example, an unusually popular menu or a special event might distort normal dining behavior. Our system provides continuous, automated measurement, eliminating these gaps and giving operators a clear picture of real‑world waste trends. As a result, sustainability teams can focus on reducing waste instead of collecting data.
Kertai: What is your vision for Clean Plate Innovations future?
Oldak: We want to become a long‑term technology partner for food‑service organizations by building custom data tools that address their specific operational challenges. The insights from our system already help operators adjust portion sizes, redesign menus, and refine purchasing decisions, and many see measurable reductions in food waste within the first few months of deployment. Ultimately, our goal is to help organizations operate more efficiently, reduce food waste, and meet their sustainability goals by making data‑driven decisions easier to act on across the food‑service industry.
