Extracting and analysing unstructured data from thousands of pages, documents, websites, and other sources has become a tough task for banks, financial institutions, insurance companies and rating agencies worldwide. Most AI automation companies claim to have solutions for rapid extraction of unstructured data, which works well for simple and standard format documents like invoices. However, these solutions fail to extract accurate data from complex documents like legal agreements, contracts, or company financials.
nRoad, a document automation platform built with deep learning technology, is pushing past the traditional boundaries of document management. Founded by Aashish Mehta (CEO), Hrishikesh Rajpathak (CTO) and Prabodh Sunkara (COO) in 2021, nRoad has trained its AI models on a vast amount of data from financial and non-financial documents, to generate insights and causal analysis using Natural language generation (NLG) and custom outputs per customer’s needs.
In the beginning…
Born and brought up in Pune, Rajpathak has done his BE Electronics from Sinhagad College of Engineering and Masters from ISquareIT. He then joined Persistent Systems and worked there for five years.
Rajpathak said, “I worked on multiple different technologies till 2011 while I worked at Persistent. I had been to the USA at the client’s side for a shorter period, and then Persistent started my L1 visa processing. Meanwhile, I was interested in doing my own startup, but back then there was no buzz around startups in India.”
“The notions that you need a lot of capital to start a business or you need to be from a business family to be successful as an entrepreneur, those concepts had started getting broken down. I discussed with my wife that if we move to the US, we will get used to a comfortable lifestyle and I will always have this feeling that I never tried entrepreneurship. She supported me and then I quit my job to start my own business,” he said.
{{/usCountry}}“The notions that you need a lot of capital to start a business or you need to be from a business family to be successful as an entrepreneur, those concepts had started getting broken down. I discussed with my wife that if we move to the US, we will get used to a comfortable lifestyle and I will always have this feeling that I never tried entrepreneurship. She supported me and then I quit my job to start my own business,” he said.
{{/usCountry}}Early investor
“People usually quit their jobs when they have an idea to work on passionately. I made a ‘mistake’ that I just quit my job because I had to do something, but had no specific idea of what to do, recalls Rajpathak. He said, “During one of the networking events, I met Mehta who was also from the field of Natural Language Processing, and we both discussed some ideas. We started formulating something very generic and later decided to develop my first product for social media analytics. He became my early investor.”
“It was like a single-person company. I never knew how to do sales, but continued developing the product. I tried my first effort at sales, but with little success. Mehta continued to support me throughout that phase. Later, we sold the product for a small cost, which I really cannot call an ‘exit’,” says Rajpathak.
Initial venture
After the first entrepreneurial ‘failure,’ Rajpathak came up with the idea of developing a food rating app. The food discovery space was starting up then. Several apps were helping users identify restaurants to eat at, but there was no mechanism to rate the food items at those restaurants. With Mehta continuing as an investor and Sahil Khan as co-founder, Rajpathak founded Quinto.ai in 2015. Rajpathak was hopeful about Quinto being successful as they were able to get a few people to talk about it. The entrepreneur duo also raised their first-ever official funding round from Jaydeep Barman of Faasos.
“B2C is a very tough game to crack and we realised this as we could not scale up significantly. Over a period, funds started drying up and a few employees also left. At Quinto, we had developed tech, where people could search for something and ask questions in their natural language. The app would work like a virtual foodie friend who knows your preferences and with whom you can chat any time. In those days, a lot of B2C products were moving towards chat-based customer support. We realised we have the NLP tech. Businesses looking for automated customer support were interested in our NLP tech and hence we decided to pivot,” he stated.
Quinto went on from being a food app to a B2B SaaS product that businesses automate their customer support over chat. Quinto was later acquired by Netcore Solutions, a Mumbai-based marketing technology company. Rajpathak served as the Chief Data Scientist at Netcore for the next 2.5 years.
Investor turns co-founder
Meanwhile, Mehta’s own company RAGE Frameworks was acquired by Genpact in 2017. The AI (NLP, NLU) based solutions developed by Mehta had transformed financial statement spreading, lending and wealth performance reporting functions within some of the largest banks and financial Institutions. He continued his role of driving digital transformation within banks financial services and institutions through Genpact’s AI solutions. He was at the end of his lock-in period when he asked Rajpathak about his next plans. Mehta had a concept of automated document processing along with a good understanding of the problem statement and market. He was good at sales and hence he asked Rajpathak to join him. Mehta along with his colleague Prabodh and Rajpathak started nRoad in 2021.
nRoad
“A lot of data today is in unstructured form. From websites to documents, data is not completely structured for a machine to analyse. The problem is how we consume this unstructured data and how we get insights out of it. nRoad processes this kind of unstructured data and can gather insights from it. Whether it is about analysing a 500-page annual report of a public listed company in the US and extracting specific data points out of that, whether they are in paragraphs, tables, or in infographics, nRoad can extract the data,” explains Rajpathak .
“From analysing big listed multibillion dollar corporates for risk analysis to analysing bank statements of individuals to understand the risk associated with personal loans or credit card defaults, nRoad helps banks, financial institutions, insurance companies and rating agencies to analyse unstructured data for any company,” stated Rajpathak.
Deep learning
Making it simpler for the readers, Rajpathak explains, “There are tables, paragraphs, infographics, graphs, and other forms of representing data in the documents. With this kind of unstructured data, we must first identify the different types of data in a document. After identifying data types, we must develop different strategies to extract this data and analyse it further. The way to extract data out of tables is not the same as that for paragraphs. We use multiple proprietary machine learning and deep learning techniques for this purpose. There are vision-based and NLP algorithms on both deep learning and statistical machine learning side. This process happens through multiple engines that are targeted towards specific types of data representation and everything is put together to solve for an entire document.”
Onboarding
Explaining the client onboarding process Rajpathak said, “Once client communication is established, we try to understand the right kind of problem statement that we can solve for the client. Then we start interacting with the customers. We give one or multiple demos after which queries are answered over a period. Since enterprise customers have multiple decision-makers, this process takes time. Once both sides are convinced that there is a problem that can be solved by nRoad we step ahead. We do a small (unpaid) or large (paid) proof-of-concept (POC) so that we get to work on actual client data. It also establishes that our engines can work on client data and solve their problems.”
“Once POC is successful, then commercials are discussed on the basis of the cost of the engine and other factors like customisations involved for the client. There are licensing and customisation fees involved with other elements like post-sales, product, and tech support, etc. A few customers also ask to update the product as their business evolves,” said Rajpathak.
Expansion plans
Commenting on nRoad’s expansion plans, Rajpathak pointed out that there is still a significantly huge untapped market in the US itself. We are looking at adjacent sectors like legal documents, healthcare, and pharmaceuticals, and other geographies like Europe and India also remarked Rajpathak.
“There are new technologies that are coming and we are constantly on our toes to implement those new technologies. We have advanced deep learning techniques and vision algorithms for large language models (LLM). We have worked with open-source language models and trained them on our own datasets so that they work well on financial documents. Our goal is to keep improving our tech,” he said.