It is not often that you hear about Indian military veterans starting a data science startup in Bangalore. With a combined experience of four decades in the Indian army, co-founders Shashi Kiran BP, CTO and Naveen Xavier, COO decided to put their military and technical expertise to good use with DataVal Analytics, founded in 2016.

Helmed by technologist and telecom guru Dr Sam Pitroda, Chairman & Co-Founder, DataVal recently made news worldwide when it aced Facebook’s famed AI task(20-part QA bAbi tasks) with a 100 percent accuracy, a first for any company across the world. Hosted by the Menlo Park giant, the Facebook AI Research (FAIR), benchmark test was created in 2015 to promote the goal of automatic text understanding and reasoning. And so far, none of the companies had achieved a 100 percent accuracy in the benchmark test.

DataVal Analytics — Solving Language Problem using AI

The startup has created an NLU engine that helps machines understand natural language. Natural Language Understanding (NLU), is essentially the computer’s ability to understand human language, and has the potential for ground-breaking solutions in areas such as the analysis of unstructured text and multilingual chatbots for example. According to Dr Pitroda, “Natural Language Understanding (NLU) capability is of significant importance for building the next generation of applications”.  

CTO Shashi Kiran BP reveals that the startup has a more nuanced approach towards NLU. DataVal has developed a new NLU technology with a deep focus on the human way of understanding language. “When we started looking at other products, we realized we were nowhere near understanding natural language. That’s when we decided to build the NLU product and we did a lot of research,” said Shashi Kiran BP.

“When you see how a human understands natural language, it is primarily based on the understanding of the world. There is a requirement of teaching a machine what  is the cause and effect of every action  and the relation between different objects in the world. A lot of common sense is involved in processing natural language,” he shared.

The NLU core engine has integrated multiple processes related to language pre-processing, word sense -disambiguation, conjunction processing, preposition association, co-reference resolution and time & space analysis, he said.

So how does this Bangalore and Chicago-based startup help machines make sense of natural language? “Since human language is largely grounded in experience and understanding of the concepts of the world, we have to start by teaching the machines the concepts of the world. Since language is created from the need to motivate an action in the world, the next step is teaching the machines what happens when an action occurs and what is the effect of that action. We are also  developing a layer of intelligence which will use this information to perform reasoning” he added.

Why Is NLU a Tough Nut To Crack?

According to Stanford Computer Science and Statistics Professor Percy Liang and NLP expert, NLU can be divvied up into four distinct categories:

1) Distributional

2) Frame-based

3) Model-theoretical

4) Interactive learning

Liang emphasizes that when the models are trained only on large bodies of text, and not on real-world representations, statistical methods for NLU usually end up lacking the real understanding of what words essentially mean.

Real World Applications of NLU @DataVal

The startup is working on two real-world applications — the first one is to build natural language queries into a structured database. The 16-member data science team is working on a large patients’ database (the startup obtained an anonymized patient database from a hospital with relevant information regarding age, gender and related diseases). “The idea is to make the database more conversational and we want the doctors to interact in a more natural way, in other words, without touching the laptop,” he shared.  

Secondly, the DataVal team is working on an application for the military which can convert intelligence reports that are in natural language into a structured form. During the span of their career, the founders dealt with technology at every stage and are now applying their learnings into building bleeding-edge applications for the military.

Bulk Of Business Comes From Data Science

However, since the NLU technology is still in a fledgling stage, the core of the business comes from Data Science projects. The startup provides Data Science-as-a-Service to a wide spectrum of industries, from e-commerce to agriculture and even commodity price prediction.

Citing a use case, Shashi Kiran shares:

Measuring Merchant Churn Rate For A Mobile eCommerce Company: When a mobile ecommerce company wanted to understand their merchants’ behavior and their churn rate, they turned to DataVal Analytics for a solution. “What happens when a company acquires large number of merchants but they face a high churn rate. The company spent a lot of money in acquiring merchants and wanted simple rules to identify which merchants will stay,” he said.

Solution: DataVal cranked out a lot of analysis and identified three simple rules to predict churn behavior, each rule having two parameters. The application was used by the agents on field who were now able to predict merchant behavior. This enabled the mobile ecommerce company to a) reorient their scarce resources on people they knew will churn b) gave them a pattern wherein they could predict the kind of merchants who will stay. They started aligning their acquiring process of merchants based on these insights.

DataVal has also worked in helping a startup from the Bay Area to build the complete framework and data science models for agriculture yield prediction using satellite imagery.

Hire & Train Model

Given data science’s talent crunch problem, many startups have deployed the hire & train model that allows them to tap into a steady stream of young talent with 1-2 years of experience. “We usually go for freshers, probably with 1-2 years of experience and spend a lot of time and resources on training them. So far, we have seen 0 percent attrition rate,” he emphasized.

Currently, the startup has 16 engineers on the team and there are plans for hiring as well.