Inarix is transforming the agricultural industry by turning smartphones into pocket laboratories for the world’s 600 million farmers. Their innovative approach enables farmers to assess crop value in real time through image-based crop qualification, leveraging the power of AI and machine learning. Inarix’s platform is built on top of our DINOv2 model, which helps their technology analyze visual data provided by customers, after previously employing the DINO model first launched in 2021. The team’s ultimate goal is to transform the entire agricultural industry, streamlining processes and enhancing crop yields worldwide.

“We are working at the frontier between AI and agriculture,” says Inarix Chief Technology Officer Artemis Llamosi. “It’s our mission to get the best out of each grain.”

Traditional crop qualification methods can result in significant waste when farmers are unable to accurately assess the quality of their grains, leading to difficulties in pricing them. Inarix’s app, PocketLab, offers a solution to this problem by allowing farmers to take a photo of their grains and analyze their quality based on various criteria, providing a detailed evaluation of individual grain properties. This innovative approach helps farmers more accurately price their grains, improving their overall profitability.

As Inarix developed the technology that drives their app, they worked closely with the Meta FAIR team to build their foundational model on top of DINO. Through this collaboration, Inarix was able to adapt DINO to agricultural data, while also providing key insights to FAIR to consider for future developments of the DINO model.

The Inarix database now contains over a billion objects. The data is constantly evolving due to factors such as climate events and differences between countries. Having a foundational model with preexisting knowledge of crops has proven beneficial, since they dramatically reduce data requirements and training time while significantly improving performance and knowledge transfer across different models of smartphones. This enables Inarix to easily learn additional criteria and expand their user base, ultimately reducing the cost of building products over time.

“It’s much faster to iterate and experiment, and it’s also much easier to learn continuously,” says Alexandre Hannebelle, Inarix’s Head of Data Science.

The company has gained significant traction in the agricultural industry, with over 600 silos across more than 10 countries in Europe and North America using its technology at an industrial scale. Inarix continues to expand its capabilities by adding new analysis criteria to PocketLab. Currently, the platform can evaluate barley, wheat, and corn, with plans to include soybeans in the near future. As a Paris-based company, Inarix now covers approximately 40% of the French barley industry, demonstrating its growing influence in the market. The product’s mainstream acceptance is a testament to the increasing recognition of the value of highly accurate computer vision in agriculture.

The adoption of DINOv2 led to a significant improvement in precision for Inarix. Previously, they were able to recognize dominant varieties and large mixes of two or three different grain varieties, but with DINOv2, they achieved precision within 2% to 3% accuracy in predicting composition—a level of specificity that informs impactful operational decisions and drives business outcomes. This enhanced accuracy led to their barley variety analysis method being officially certified by Incograin (Syndicat de Paris), marking the first time an AI software solution has been recognized by one of France’s oldest professional unions. As a result, Llamosi says farmers can now rely on proven technology to guarantee varietal purity of their crops in record time.

“Our vision is to spread both upstream and downstream, and DINO is an important element of this strategy,” Llamosi says. “Given this new lens that we provide, we think we can really transform the agricultural supply chain.”