In the first of a series of startup stories from the Product Labs stable, we take a look at a sophisticated Natural Language Processing (NLP) technology that has not only been successfully productized but also gained some high profile customers in less than a year. And it all started with an internship. Read On.

The International Institute of Information Technology, Hyderabad (IIITH) has always prided itself on having a strong research focus. So much so that it is probably one of the few tech institutes in India to actively encourage undergraduates to dabble in research. However the intent is not only to give a fillip to research but also to simultaneously facilitate the commercialization of the developed technologies. “The idea is that it should be put to good use – commercial or social”, says Prakash Yalla, who heads Product Labs, an initiative of the Technology Transfer Office at IIITH. For this, Product Labs takes research out of the labs, builds specific use cases, and translates them into solutions comprehensible to industry. The marketable solution could be a viable prototype, which if well received by industry, is licensed to them through the classical route or hived off to a startup.

Entrepreneur-In-Residence

“At IIITH, our focus has been on trying to create startups,” affirms Prakash. The primary vehicle for doing this is by leveraging their flagship programme known as the Entrepreneur-in-Residence (EIR). EIRs could be fresh graduates or seasoned professionals between jobs who are mentored by Product Labs over a period of a year so that upon completion, they are business ready and have a couple of clients too.

The Tech

The very first use case the Product Labs team tried to build was for a sophisticated Natural Language Processing (NLP) application. The underlying idea of the research itself was to enable a user to conduct an understanding-driven search as naturally as possible. According to Dr. Manish Shrivastava, Professor at the Language Technologies Research Centre and founder of the NLP startup, a ‘document expert’ like this is more complex than a mere Q&A interface. ”Here, the AI not only extracts the relevant information across documents but also understands the context in which the information is presented and simultaneously points out the location in the document where it is mentioned, adding to its credibility,” he says.

The Product

Typically, Product Labs scouts for an industry partner or an entrepreneur to build a business out the tech idea. In this case however, it was an intern assigned to building a proof of concept for this technology who decided to start up and create a product out of it. “I think what gravitated me to the problem was the fact that it was so widespread. People were very unaware of tech like this, tech that can read your documents for you and then help you find relevant answers through conversation,” says Vishnu Ramesh, who is now co-founder of the startup along with Prof Srivastava. Due to the conversational style of interaction which is in the form of a text-based Q&A, the product was initially referred to as a document bot or a docbot. However thanks to the nuanced data extraction of relevant information that is often locked up in unstructured forms, it was officially christened to Subtl. “The cognitive intelligence helps you do more with less effort; it’s trying to be subtle,” explains Vishnu.

Customer Connect

Since the technology itself is based on a deeper level of language understanding rather than a domain specific or format specific implementation, the product is generic enough to be used for various kinds of documents. Speaking of how initial client acquisition took place, Vishnu says, “IIIT was an early-stage channel partner for us. Enterprises came to IIIT with problems that our technology could solve, so IIIT got us involved in these engagements.” Subtl thus has a diverse clutch of clients ranging from the banking sector, to a large shipping and manufacturing company and even an organization involved in the procurement of defence equipment. With the basic idea being to reduce the cognitive load of an agent on the ground, in each of these cases, retrieval of specific information such as policy documents, operational manuals or even circulars and memos takes place by mimicking the presence of an expert.

CIE/Product Labs Backing

For any fledging company, getting visibility and the initial connect to the client base makes all the difference. In the case of Subtl, this difference was made by Product Labs. “I can safely say that without Product Labs, we would not have been a product or a company,” says Manish. Prakash seems to agree when he says, “We made a lot of things very easy and conducive for an entrepreneur to take off on his journey, whether it is access to tech, access to faculty, or to market. If an entrepreneur were to approach a public sector bank on his own, he or she wouldn’t have got it. Because it was backed by a IIIT brand, it made things a little easy and the cycle time was much lesser for them to close those deals.”

Challenges

Taking the entrepreneurial plunge without its fair share of hiccups is an oxymoron. However, Manish is quick to point out how the doubts faced by him in the early stages were mitigated by the presence of Product Labs. “For a young entrepreneur it is very difficult to know where to focus when faced with pressure. But due to Product Labs, the entire team was well guided in terms of where to focus their energies, and what can be taken up by other members if the need arises.” From the Product Labs’ point of view, the challenges are different. Since the EIR programme typically seeks out young students or professionals, it’s difficult to gauge their level of seriousness and intent to embark on the entrepreneurial journey. The other hurdle is generic in terms of user expectations of deep tech. “There are a lot of perceptions and misconceptions about what AI can do. Everybody wants an AI-enabled product with unrealistic expectations!,” exclaims Prakash. For instance, in the case of Subtl, they are frequently questioned about its ability to talk much like an Alexa. While a voice-enabled feature for the product would be highly desirable and is indeed in the pipeline, it is not present today thanks to the complexities in accent-specific training of voice data specially in the Indian context. “At the moment, we are sticking to text-based search and retrieval. As Indian language recognition becomes more mature, that will be integrated too,” says Manish.

Looking Ahead

With the EIR programme coming to an end, Subtl has emerged with a strong technical base, high profile customers and a lot of visibility. Because of the technology’s general applicability, other sectors such as legal firms and the pharma industry too have evinced interest in the tech. In addition to this, startups and companies using traditional chatbot interfaces who have realised that much of their customer support activities need access to docs are looking at licensing the technology. The team that had modest beginnings with 2 members is now steadily expanding. In April this year, they were selected among other startups as the result of UNPITCH, an unconventional startup pitch event that was launched by CIE specifically to help startups tide over the economic slowdown caused by the lockdown. Presently, revenue earned from their client base takes care of all their current operations. That said, the team is in the process of preparing for a significant fund raise and is poised for growth.

To learn more about the EIR programme, click here.

For more about Subtl.ai, click here