India’s chatbots pivot for survival in the age of Siri and Alexa

Photo credit: Валерий Качаев.

If you’re a smartphone user, AI has already entered your life. Most probably through chatbots or virtual assistants – Apple’s Siri, Amazon’s Alexa, Google’s Assistant, Microsoft’s Cortana, Facebook’s M, or any of the countless Magic clones and other AI helpers. The tech titans are on a landgrab to build AI ecosystems – going beyond the boundaries of devices like iPhone, Echo, or Pixel.

But in India, the virtual assistant war has been playing a slightly different tune, using a combination of AI and human help on text-based chat. Chatbots here are trying to iterate and find survival strategies to avoid the fate of many of their counterparts in Southeast Asia who choked to death due to meek adoption, failed transactions, and a money crunch.

Some are integrating with big brands for monetization, some are trying premium assisted services on a subscription model, and some have become bot-builders. Now, one of the oldest of them has made a key component of its chatbot tech open source in a bid to hook developers and users.

Today, India’s first chat-based virtual assistant, Haptik, threw open one of its proprietary tech components – Named Entity Recognition​ (NER) System – that helps its bots decipher conversations better. For example, if you say: Send Pehu one Ruby Woo lipstick from Mac tomorrow, the AI powering the assistant should detect the name of the entities in the block of text and annotate it. So with NER, the earlier command would be annotated quite like this: Send [Pehu]{Person} one [Ruby Woo]{Product} lipstick from [Mac]{Brand} [tomorrow]{Time}.

The response to your request will depend on the strength of the NER; in other words, on how well the system recognizes these key words for what they are.

Haptik, which has been working on it since 2013 – well before chatbots were cool – now tackles over 5 million chatbot requests in a month, its co-founder and CEO Aakrit Vaish tells me, suggesting that its AI has built some serious tech muscle.

GIF source: GIPHY

Why open source?

“If you have to build NER from scratch, it would take years of data mining. Instead, developers can simply download, install, customize, and use our NER for free to build and power their chatbots,” he says. “Be it for personal assistance, ecommerce, insurance, healthcare, fitness, and so on. Any tech product that runs machine learning will need NER.”

Vaish founded Haptik with Swapan Rajdev in 2013, and launched the app in 2014. Like YC company Magic, AI and real human beings help Haptik get routine jobs done for its users on chat.

Haptik’s NER code is hosted in this Git repository. “Any developer can pick a chatbot-builder like API.ai or Gupshup, start building their own chatbots, and cut short their development time by using our NER plugin to detect all the keywords,” Vaish explains.

“We are the first company anywhere in the world to open source a key part of chatbot tech, specifically this NER piece,” he adds.

So, why open source such a vital cog and lend muscle to potential rivals? “There’s a lot going on with chatbots but it is still very much Day 1. Developer tools and open source technology play a key part in the evolution of any platform,” Vaish says. “Our broader goal is to make chatbots ubiquitous. We believe that everybody in the world, at some point, would be using chatbots to get things done.”

The starving Magic clones

Where’s the money?