Now, you can use artificial intelligence to pick a Bollywood saree

You see a dress you like on a fashion portal and want to compare it with similar ones to make a choice. It seems simple. But do you realize how smart your brain has to be to decipher that one dress is similar to another? Color, light, dimensions, depth, gender – your brain processes a multitude of inputs to make the call.

Now, imagine trying to get a machine to do that. To come up with intelligent suggestions for dresses that match the one you like, a computer has to convert images into numerical and symbolic data. This is a complex task involving geometry, probabilities, learning theory, and so on. Of course, once it learns to do that, it can scan a lot more dresses much quicker than we humans can, so we get better choices.

Tagging to seeing

Indian fashion portals Voonik and Craftsvilla are rolling this out today: if you go to these sites and click on a Bollywood saree or terracotta jewelry you like, artificial intelligence will kick in and recommend similar sarees and jewelry to you instantly. (For a couple of weeks, the feature will be in A/B testing mode, which means only half the users can access it.)

We’ve got used to seeing recommendations for products when we click on books, apparel, or anything else on ecommerce sites. But those are different from visual comparison. “Earlier, we only had recommendations based on tags. We look at the physical attributes of items (color, sleeve, fabric, etc.) and suggest items which have similar attributes. With the new feature, we show items that are visually similar to the item the user is viewing,” explains Sujayath Ali, co-founder and CEO of Voonik.

Tags are text-based descriptors which may even include visual attributes like color. But they don’t tell you the exact shade of color, pattern, or shape of the dress like computer vision can. That’s why you often find oddball recommended products. After all, describing an object with tags cannot mimic how we see an object in its entirety. But computer vision comes close to seeing an object as the human brain does.

Apart from tags, recommender systems on ecommerce sites rely on user data – such as, what you’ve liked earlier or what’s popular among other users. But for fashion portals like Voonik and Craftsvilla, in particular, it makes sense for the recommendations of similar products to be based on the visual attributes of something you click.

AI on the cloud

Ashwini Asokan and Anand Chandrasekaran, co-founders of Mad Street Den

Behind this visual recommendation engine is a cool product called MADstack – a cloud-based suite of computer vision and AI modules that an ecommerce or any other site can plug into with an API (application program interface) to create new experiences for customers.

MAD stands for Mind-Abled Devices, and MADstack is a creation of Mad Street Den, an Indian startup that’s barely two years old. Its founders, however, are seasoned pros from the hallowed domains of AI, user interface, and neuroscience in Silicon Valley.

Ashwini Asokan led the mobile innovation team at Intel’s Interaction and Experience Research Lab. Her husband, and Mad Street Den co-founder, Anand Chandrasekaran was part of an AI team at Stanford University that built Neurogrid – a device that simulates the neurons and synapses of the brain to control bionic limbs.