Sorting potatoes by hand is physically demanding and time-consuming. This is exactly where Karevo comes in. The sorting machine of the start-up founded in 2024 uses a specially trained AI model to automatically inspect potatoes.

For training, more than 100,000 images were used. According to the start-up, the model can identify damage to potatoes with an accuracy of 95 percent . In addition to foreign bodies, the system recognizes seven different defect types, including rot, cracks, and damage caused by wireworms.

An important aspect for use in agriculture: the AI can also analyze unwashed potatoes . This is technically challenging because their appearance can vary greatly depending on the region, soil type, and storage conditions. Karevo's AI model makes it possible to adjust the sorting to the specific potatoes of a farm.

Karevo specifically targets its technology to the requirements of small and family farms. According to the founder's experience, existing potato sorting machines are often too large, expensive, or maintenance-intensive for smaller operations.

The Karevo machines are therefore smaller and modular in design. This is intended to make them easier to maintain and integrate into existing systems.

From master's thesis to AI start-up

The development of the technology began during Keßler's master's thesis at the Technical University of Munich (TUM). Together with engineering student Johannes von Wittke from the Ostbayerische Technische Hochschule Regensburg, he further developed the founding idea.

Through UnternehmerTUM, the center for innovation and entrepreneurship at TUM, Felix Beck joined the team, who completed an Executive MBA at TUM. During the founding phase, the team participated in the UnternehmerTUM Incubator and developed the first prototypes in the UnternehmerTUM Makerspace.

The team founded Karevo in 2024. The company has been on the market with its technology since autumn 2025.

AI for practical automation in agriculture

The Karevo example shows how computer vision and specially trained AI models can address very specific challenges in agriculture. Instead of using a general AI solution, the system is specialized in a clearly defined task: automatically recognizing potatoes under real operating conditions and sorting them according to quality characteristics.

Thus, the start-up combines AI-based image recognition with mechanical engineering and an application case that directly arose from agricultural practice.