Ankit Narayan Singh of ParallelDots
Now a team of 15 engineers, ParallelDots are immersed in building machine learning systems and are excited about their applications in both text and image processing. Ankit shares their earliest enterprise applications:
“We began working with online publishers to create timelines for their past content. For instance, if we had worked with a news website that is covering Brexit, we could quickly gather all referendums from the past in a timeline to give a fully immersive, comprehensive backstory”. Ankit elaborates: “Our timeline generation tool was completely automated and at our peak, we were serving around 100 million page views a month with an accuracy better than any tools used by publishers. We cracked the semantic search technology code and deployed it at scale”.
But working directly with publishers was only the start.
“We quickly learned that our technologies have broad application beyond publishing. So we decided to expose our proprietary technology stack as APIs for anyone to build smart applications and products”.
Capitalising on their existing expertise, Ankit explains how ParallelDots built an expanding set of tools for any enterprise to access the machine learning technology and integrate machine learning capabilities into their own applications.
“We began developing several machine learning APIs and outsourcing them to companies for their own use in-house. Fortunately, in the last six months we have become increasingly trusted by other enterprises using our APIs to analyse different types of data sourced from verticals like e-commerce, news, social media, real estate etc. For example, while our text analysis product is still being used to create timelines, using our Semantic Analysis API now enables companies to do it themselves more quickly, easily and on a larger scale”.
Ankit tells me that they are currently working with Newsbytes, an Indian news app that aggregates Indian and international news.
“The software can scour through past news – all the way back to 2002 – and recognise patterns in similar events from the past. So, for example, if the Indian prime minister has made a controversial decision, our algorithm will index other similar results from similarly controversial decisions made in the past to create a timeline automatically”.
ParallelDots have recently found a particularly exciting use for another API, the Keyword Extractor, which helps brands collect feedback from across the web.
“We have now integrated a combination of the Keyword Extractor and Sentiment Analysis API with a video analytics start-up to get feedback on comments posted by users on various video hosting platforms to the biggest brands. For instance, when Apple or Samsung releases the video of a new smartphone on YouTube they can get comprehensive, actionable insights into the reception of the video within minutes. This extends to getting insights not only on their product but also on their competitor’s products, to ensure their offering is in tune with what’s being offered by competitors and demanded by end users”.
This also crosses over on to Facebook and Twitter commentary, where comments on brands’ Facebook pages and Twitter profiles also get the same deep level of analysis.
“These social media sites are key sources of feedback. Twitter more than Facebook in fact, as Twitter is far more public on the whole”.
Very recently, ParallelDots applied their Sentiment Analysis and Entity Extraction API to the Euro 2016 finals hashtag #Euro2016 to gauge fans’ positivity percentage about their teams and players. You can try out the Sentiment Analysis tool yourself.
ParallelDots are already working with enterprises and other start-ups, but as a start-up themselves they have come a long way in a short time. So where are they headed next?
Ankit remains open-minded at this early stage: “We are building products in multiple verticals – the common thread is the technology platform on which we have multiple teams standing. Our prime focus is to have the strongest world-leading level technologies in AI/ML, but we don’t deploy technology; rather, we build products that deliver profitable outcomes for our end users”.
Their strength in the market is already evident, he says: “We have been providing technology services to other companies who have found us without us proactively reaching them. They find us based on our reviews and studies we publish on various forums. We did a study on the automobile industry, analysing posts around a particular brand of new car. One item we demonstrated was if an issue came up on a manufacturer’s Facebook Page repeatedly through the comments section, it would be known immediately and could be acted upon straight away, rather than the delays and disconnects that happen in more traditional mechanisms of collecting feedback from service centres and then acting upon it”.
AI and ML uptake is increasing exponentially across the enterprise, but there are still factors that are getting in the way of mass adoption. We asked Ankit for this thoughts on the main obstacles:
“The first and foremost challenge we know about is broadly data-related”, he says. “Enterprises generate a lot of data but sometime they struggle to figure out what data to analyse, which tools to use, and how to measure the ROI of their analysis. Then, there is issue of data security and privacy and sometime they’re not sure whom to trust with their data. We have experienced this first-hand many times in our deal negotiations where use and protection of data often become the hardest topics to figure out.
The next step for ParallelDots is integrating with Instagram APIs, and building upon existing visual recognition and natural language capabilities to build smart leading edge applications and products. The company is also building computer vision capabilities to tackle image data, and while early test deployments of image processing systems have started, things are still under wraps. Ankit’s comment on image processing is “fasten your seatbelts”.
Happily for Ankit, clients are already report exciting results, and he is assured their systems will get “much better”.