For most shoppers, it is easy to spot the squelchy, sad-looking tomatoes marooned on the back of the shelf. Subtle traits in color, feel, smell, or texture swiftly drive reactions like, “ah, well: they’re discounted and still edible; I guess I’ll still buy them,” or, “it’s definitely past its best, so I’ll find another batch somewhere else,” or even, “those tomatoes are so awful, you can’t even throw them at a medieval criminal languishing in the stocks; this is the last time I shop here.”
Judgements like these become exhausting and flawed, however, if you have to make these sorts of subjective assessments to a few hundred boxes of tomatoes, let alone assessments by the tens of thousands. Yet this is exactly what retailers have to do, and they are hardly helped by the length and complexity of the world’s fresh produce supply chains. Get the judgement of tomato freshness entirely correct one day? Those findings, as is the way with perishable goods, could be wrong by the next! So quality control checks often have to be repeated across journeys that can span hundreds of miles — at each handover, from producers to a whole string of middlemen or inspectors before it finally gets onto a shop floor. All the while, consumer demand surges and wanes as expectations firm up for year-round availability of seasonal or exotic items; lead times fluctuate with variable harvests or even political brakes on exports through surprise customs checks. So the whole thing is a logistical headache, and this is why an astounding amount of it gets tossed out somewhere along the way between the farm gate and a restaurant plate. Or, the problem exists in reverse: low grade, rotting food like those tomatoes creep onto the shelves — which as a recent McKinsey report points out, can be highly detrimental to the brand perception of an entire supermarket chain.
A startup addressing this problem with an initial focus on the Asia-Pacific region is an Indian B2B agritech software company, Intello Labs — a member of AgFunder’s Singapore-based GROW Accelerator, where it was part of the first GROWhort late last year.
Too many of today’s food quality control checks are in themselves still heavily manual, time-consuming, labor-intensive, and error-prone across much of the world, Tanmay Bhargava of Intello Labs tells AFN. The result: these errors in procurement or transport are hard to trace; there could be a quality checker who has gone rogue at any number of points along the way — perhaps simply erring on the side of caution, or just working off a subjective, inconsistent or vague set of standards for what constitutes “fresh.” Worse, a human may never have even seen a rotting fruit with their own eyes; manual checking, Bhargava explains, is usually done through taking a sample of just a small portion of the entire inventory as a way to save time.
To find a solution, Intello is turning to deep learning and computer vision — can you effectively train computers to accurately spot and flag when food is spoiled with greater speed and accuracy than people? (And note: the traits of a fruit or vegetable rotting are diverse, making this a harder ball game to train for than equivalent technologies that identify tomatoes that have ripened and need to be picked — if red, then pick; if green, leave alone.)
Headquartered in Gurugram, India, with a presence in Singapore and New York, Intello has four co-founders: Milan Sharma, Nishant Mishra, Himani Shah and Devendra Chandani. The first three had known each other closely for more than a decade, since their undergraduate days at the Indian Institute of Technology Bombay. Prior to founding Intello, the company’s CTO Nishant worked on AI and computer vision for five years with companies like Canon and Amazon.
“Our image-based and hyperspectral products bring transparency and standardization to quality assessment of fresh commodities, mainly fruits and vegetables. We help reduce value risk and wastage in agriculture supply chains. Images of food items, whether they are taken from an employee’s smartphone or by a mounted camera, are processed by our cloud-based AI to generate almost-instantaneous quality reports and actionable insights,” is how the Intello Labs team describe their technology. “Our algorithms identify each unit of the food commodity visible in the image and classify it on the basis of size, color, shape, health and defects. The results of all the units are then aggregated to calculate the final quality grade of the lot. The whole process takes anywhere from a few seconds to minutes, depending upon the commodity and the sample size.”
The company says it enables food retailers “to measure, monitor and control quality as per customer expectations through non-destructive and scalable methods across the supply chain. They use our versatile products, which are implementable at vendor sites, procurement centres, warehouses, distribution centres, retail display.”
Its value proposition can be split into four columns:
If any of this is familiar to readers of AFN, it is probably because there are a handful of startups, including India-based Agricx and San Francisco-based AgShift, that offer similar products and services. Agricx’s product is designed mainly for potato and tea leaves, and requires a controlled environment and careful sampling, while Agshift’s utilises large format hardware to maintain a controlled environment.
Many larger players, such as Tomra, Unisorting, Buhler and Forsberg, also make grading and sorting equipment, but they must combine electro-mechanical or pneumatic systems to spectroscopic or chemical analysis. Intello Labs, however, are backing their approach in terms of its cost, effectiveness, and scalability.
The team describes its single most important differentiating feature as scalability and the speed it can create a customized product for a new commodity. “In order to scale up faster across commodities, we have developed a self-learning internal platform that is versatile in its learning capabilities. This is our key differentiator. With this platform, our development time for new commodities has come down to weeks and the training process has become seamless. With it, even non-technical employees can help train a model with the input data. In the near future, we plan to offer this platform as a service to customer organizations, whose employees can use their internal data to train the models inexpensively and quickly.”
IntelloLabs’ computer vision algorithms have has been trained on a large and high-quality data set of images of fruits in particular stages of spoilage that the team has gathered first hand or through their retail partners. Clientele include Reliance Fresh (India) and Walmart (China) along with a few other major international retailers in the region. “We make white-label products, which come in the form of either mobile apps or as APIs, to be used by corporations in the agriculture and food sector.”
IntelloLabs has several other large companies in the sales pipeline, and the 70-strong team is currently in the process of expanding strategically to Australia where there are high margins to be made on good quality products that have endured long supply chains on their way to market.
How would you improve food supply chains with artificial intelligence? Let us know by shooting an email to [email protected]