The team that won the 2016 US Baseball World Series relied on data that was analyzed and tagged by women in Metiabruz
The Chicago Cubs won the US Major League Baseball World Series title in 2016, its first win in 108 years. The LA Dodgers reached the 2017 World Series final, before losing in a game tainted by a cheating scandal. What the two teams shared in their dream runs was use of AI.
Florida-based Kinatrax had high-speed cameras installed at strategic points on baseball grounds for synchronized motion-capture videos of pitchers. These were annotated, tagged and analysed to create the 3D anatomical models that fine-tuned pitching mechanics for each player.
Little known in this tale of tech triumph is that a group of women in Metiabruz, a predominantly Muslim township on the fringes of Kolkata, played a crucial role in enabling the Cubs and Dodgers pitchers to outperform rivals. They work for iMerit, a startup based in Kolkata and California, that provides the data annotation vital for training AI algorithms. Kinatrax was among its growing roster of global clients. “One thing I never expected was to be mentioned in Sports Illustrated," says Radha Basu, iMerit’s founder, with a twinkle in her eye, referring to an article celebrating the Cubs’ World Series win that had a hat tip for AI. Before iMerit, she started Hewlett Packard’s software centre in Bengaluru in the late 1980s and then took Support.com to a Nasdaq IPO as its CEO.
In 2006, Basu and her husband Dipak launched a non-profit social enterprise, Anudip Foundation, to help marginalized youth and women find digital livelihoods in India’s ecommerce sector. A few years later, during a discussion on the future of work with Pierre Omidyar, founder of eBay who later became an impact investor, Basu had an epiphany: why not leapfrog into an area like data annotation which was starting to see huge global demand with the growing adoption of big data and AI? That was the genesis of iMerit, with seed money from Basu and Pierre Omidyar in 2012.
One of its first two centres was in Metiabruz, where Nawab Wajid Ali Shah created a Mini Lucknow after being ousted by the British from Awadh. Tradition made it hard for women from this area to relocate for tech jobs after training by Anudip. So they were good candidates for iMerit, whose mission was to build a revenue-making enterprise in tech with an inclusive workforce drawn mainly from non-metro areas.
First, there was a mental barrier to cross. “Who would believe that a group of young women somewhere outside Kolkata could work for the new economy? It reminded me of the late eighties when people found it hard to believe that you could do IT and Unix work from India," recalls Basu.
“We initially trained our people on tasks like product categorization and sentiment analysis on crowdsourcing platforms. They became really good at it because young people are agile, nimble and very competitive."
Starting with a group of 60-70 people, iMerit kept expanding its workforce and getting into higher value tasks like tagging and annotation for computer vision and natural language processing. Today it has more than 2,800 employees in nine centres, including one in Bhutan. More than half the employees are women and over 80%are outside metros. The average age is 24. “Di, if not for you, it would be 23, they tell me jokingly," says Basu.
The Metiabruz centre has grown from the first 30 women to 600 now who do computer vision work with clients like Kinatrax. Another centre near Kolkata, in Baruipur in South 24 Parganas district, supports financial and agricultural data analytics. A centre near Bhubaneswar specializes in medical applications of AI.
iMerit’s Ranchi centre works on images from drones and autonomous vehicles. Its Bhutan centre has been tagging aerial images to track poachers. A centre near Shillong, where proficiency in English is higher, works on natural language processing. There’s also a centre in New Orleans, which is creating jobs for people from tier-2+ towns in the US.
There are unique benefits to a tier-2 workforce. “Our attrition rate is 6% whereas it is 30-32% in India’s IT industry," says Basu. “It’s different from back office or BPO work. When people are learning new things, they stay and grow with you."
Most of the 140 team leads at iMerit also come from under-resourced communities. Barnali Paik was one of the first women to leave Karanjali village for a tech job when she joined the Baruipur centre in 2012 right after graduating with a degree in Bengali. She got promoted as a team lead and subsequently a project manager. Taking her lead, several other women from Karanjali have made their way to iMerit.
“I remember reaching the centre two hours early on my first day and being very nervous. But gradually my fears subsided as I performed well in the data work," says Paik. “Now I’m handling a project for geospatial agricultural data annotation across multiple centres. Some of my team members are in Bhutan."
Subject matter expertise grew in step with iMerit’s involvement with global clients like Microsoft, AWS, geospatial data analytics company Orbital Insight, and Sentera, which provides agricultural mapping for crop management.
Jeff Mills, iMerit’s global vice-president (sales), explains a typical trajectory: “One of our clients derives insights from geospatial data. We started small with five technicians marking assets from aerial photos. Over time, the client’s requirements expanded as their customer base grew. They needed our help to solve unique problems. Each new project had a different set of complexities. Our team grew to 50 technicians as a virtual extension of the client."
Taxonomy varies by subject. In agriculture, for instance, the many types of blight are gathered from experts. Annotators are trained to identify each type in images. “I’ll never forget this huge standing skeleton we brought from a lab six years ago because we had to explain movement of joints. We made a digital skeleton to train people in key point annotation for medical applications like prosthetics," recalls Basu. “That’s how we got into computer vision and image work that later went into analysis of motion capture videos for baseball."
She believes women are particularly adept at the nuanced precision required in annotating images for AI engines. “You could track a small object from frame to frame, distinguishing it from a number of other objects. Let’s say you’re looking at poachers, identifying them from the way they move. This is knowledge work and not rote work."
Early backing for iMerit came from Omidyar, Khosla Impact and Michael & Susan Dell Foundation. In February, CDC Group led a series B round of $20 million. This eased the transition from iMerit centres to work-from-home, which was no mean task, given the large screens that data annotators use.
Sumit Chakraberty is a consulting editor with Mint. Write to him at email@example.com
Subscribe to Mint Newsletters
* Enter a valid email
* Thank you for subscribing to our newsletter.
Never miss a story! Stay connected and informed with Mint.
our App Now!!