Business Standard

Building the blocks of artificial intelligen­ce

- NIRMALYA BEHERA

Artificial intelligen­ce (AI) might be a buzzword for modern businesses but machines are smart only when fed with data to think and act like humans. Playment, a Bengaluru-based start-up, aims to serve the purpose of providing high-quality training data, though its people work from home, for computer vision-based companies.

Founded in 2015 by Siddharth Mall, Ajinkya Malasane and Akshay Lal, all IIT graduates who have worked for Flipkart, Playment raised $1.6 million recently as part of a pre-Series-A funding from Y Combinator, SAIF Partners, Sparkland Capital and Silicon Valley angel investors Ryan Petersen (CEO, Flexport), Max Altman, among others.

“We’re building the annotation infrastruc­ture which will help a generation of AI companies train their models and bring products to the market. We do this via our managed ‘work from home’ community of 250,000-plus users, also known as players, who have downloaded our app,” said Mall.

Otherwise, data annotation­s are done by business process outsourcin­g entities for companies working on AI. This comes at a high cost. Mall says data annotation is labourinte­nsive. For companies working on autonomous driving, each hour of data collected takes almost 800 human hours to annotate. It takes about 100,000 images and a week of teaching for a computer to safely learn a traffic situation. When you drive, there are multiple decisions you need to make among so many uncertaint­ies — weather, lighting, type of vehicles, oneways, etc, he said.

With a 30-people team, the company claims to be providing the same throughput as a 1,200-seater BPO. “We have built state-of-the-art tools specifical­ly targeting companies working on AI and provide a fully managed service with enterprise grade SLAs (service level agreements).”

The start-up is betting highly on the AI market, which is predicted to grow to $47 billion in 2020, says market research firm Internatio­nal Data Corp. The competitor­s to Playment are Amazon mTurk, Crowdflowe­r and Mighty AI. “But we are significan­tly cost-effective. Further, no one else provides a fully managed service,” said Mall, who is the chief executive. The company has Flipkart, Olacabs, Paytm and the Alibaba Group as clients. There are original equipment makers, start-ups and suppliers in autonomous driving, satellite imagery, facial recognitio­n, e-commerce and optical character recognitio­n. Over half the revenue comes from global customers.

Mall says Playment has generated 50 million tags till date and the current capacity is to process over a million a day (equal to 10,000 manhours daily). The company has a custom pricing model that depends on the project scale and complexity. It also has a few annual contracts. In July 2016, Playment had also raised $700,000 from SAIF Partners. “We would be expanding our product, engineerin­g, and data science teams to build a complete suite of services for all kinds of AI use cases, improving our quality models and build an internal AI research lab. Ajinkya would be setting up an office in the Bay area, in Silicon Valley,” he said.

Bengaluru-based Playment aims to serve the purpose of providing high quality training data, though its people work from home, for computer vision-based firms

Newspapers in English

Newspapers from India