Oct 19: Agrograde, a physical A.I. startup building technologies that make fresh produce packhouses autonomous, has introduced India’s first AI-powered optical sorting system developed specifically for arecanut, with the first commercial deployment installed at Vikas Traders in Mangalore, Karnataka.

The system can process up to approximately one tonne of arecanut per hour and enables the Mangalore facility to process close to eight tonnes a day. It uses computer vision and AI to inspect individual nuts, identify defects and automatically separate rejected material from accepted produce.

The system is designed to address one of the persistent challenges in the arecanut supply chain, i.e,  maintaining consistent quality while sorting and processing large volumes of nuts. Unlike conventional colour-sorting machines that primarily identify differences in colour, Agrograde’s system uses computer vision and AI to inspect individual arecanuts for external defects that may not be distinguishable through colour alone.

“Our machines are designed after studying multiple characteristics of a crop. We study how a product rolls, glides, the residue it leaves, the amount of dust that comes with the product, the amount of force that is needed to gently sort the product, and the variations across different varieties. We also understand the current process to understand the nuances in quality assessment. We have a dedicated team that conducts this research which enables us to develop one of the best AI powered optical sorters rivaling even imported technologies. It took us almost 5 months and 3 product iterations to build India’s first autonomous optical sorter for Arecanuts. This AI sorter has made warehouse operations almost completely autonomous. We are also grateful to the team at Vikas traders for trusting us and helping us customize our technology for the Arecanut supply chain.,” said Rakeshkumar Barai, Co-Founder, Agrograde. 

The design of the machine is based on Agrograde’s Vector architecture which ensures high surface coverage when an object moves through the vision system. It also handles high variations in size and shape of produce extremely well. Aniket Ghadage, Head of Product Engineering mentioned that Agrograde’s vector architecture follows a design philosophy that makes sophisticated machines easy to operate and maintain and is designed to run even in challenging environmental conditions such as heat, dust, and humidity.

The technology was developed over several months in collaboration with Vikas Traders, a Mangalore-based agri-business with decades of experience in the arecanut trade. The collaboration was aimed at adapting the technology to the crop’s specific characteristics and the operating conditions of an actual arecanut facility, including dust, foreign matter and high humidity.

When initially commissioned, the arecanut sorting system recorded approximately 75% sorting accuracy. With more commercial data becoming available from operations, the AI models have since been updated, taking accuracy to approximately 89%, with further improvements expected as the system continues to learn from additional data

Before automation, Vikas Traders relied heavily on manual inspection and sorting. According to the company, an individual worker could manually inspect and sort roughly two bags a day, while a team of around 15 people could process approximately 40–45 bags a day. The availability of skilled manpower also influenced how quickly consignments could be inspected, packed and dispatched.

The automated system now provides a more predictable processing capacity. It also generates a quality summary after every lot, including information on size distribution, defects identified, defect intensity and the number of nuts packed, giving the business additional data to monitor quality.

Raj Padaliya from Vikas Traders said,

"I was curious to see if AI and robotics can be deployed in the arecanut industry. I was looking for technologies which could make our entire warehouse autonomous and fast. We were excited to know that there is an Indian startup working on this, and we worked with them closely to try out multiple versions. Finally, we had success and now our sorting process is almost completely autonomous. We are the first private company in the Arecanut industry to adopt such sophisticated and high-level A.I. technology. I'm amazed at the level of hardware development that Indian startups are doing."

The system works by using Agrograde’s patent pending vector technology that captures 10-12 images per nut. Agrograde’s proprietary AI models analyse the images within milliseconds and identify visible defects before the sorting mechanism separates the rejected material from the accepted lot. The models have been trained using lakhs of crop images and run on high-performance computing infrastructure powered by NVIDIA.

Agrograde has been developing AI, computer vision and physical automation solutions for Indian post-harvest operations across crops including onion, potato and apples. The company recently deployed its 150th machine across 15 Indian states. The arecanut deployment marks its expansion into a crop where quality assessment has traditionally depended heavily on manual inspection.