[Tech50] This Hyderabad-based startup will help you extract the exact info from a haystack of documents
Having to flip through a 20,000-page document to find just the relevant information is not only taxing but a sheer waste of time. One might even try the good old ‘Ctrl F’, but what if you are looking for something related to ‘assets’ in the document, which instead uses the term ‘properties’?
This is where Hyderabad-based startup YourStory’s Tech50 2021 list of most promising startups, comes in. One can ask a simple question in the most “layman language”, and the software will seamlessly process a large page file document with many thousands of pages to fish out the relevant information in a few seconds., which is among
The one-year-old startup, seeded at International Institute of Information Technology Hyderabad (IIIT-H), is the brainchild of 23-year-old engineer Vishnu Ramesh and Professor Manish Shrivastava. It was during Vishnu’s research internship under Professor Shrivastava at IIIT-H, when the duo decided to use their synergies to devise a technology to fill in the gap.
“The question we asked ourselves was—why does anyone have to open a thousand page document to find the information they need. Time is money in today’s world and the pool of information is infinite,” says Vishnu, alumnus of Birla Institute of Technology and Science (BITS), Pilani, Hyderabad campus.
The head start
By 2019, the founding duo had managed to curate a prototype and spent another year building on top of it before formally launching their startup in January 2020. During their prototype phase, the State bank of India (SBI), which holds an “affinity” with IIIT-H, was looking to innovate and find a smarter way to streamline their processes. The startup grabbed the opportunity and partnered with the State Bank Institute of Innovation & Technology (SBIIT) to present their innovation and explore the technology.
“The bank has over two lakh employees and the information access for each of them is very fragmented. There are way too many documents and one has to manually go through the entire portal to pull out information. The other option for them is to reach out to the subject matter expert or a peer which has its own delays. The idea was to streamline all the processes for SBI, reduce peer dependency and have a quick turnaround of information,” explains Vishnu.
Subtl.ai managed to get its first purchase order (PO) from SBI three months before formally launching their startup. “The partnership further convinced us of the value of what we have built and it’s scalability,” he says.
To test the proficiency further, the startup conducted an internal testing with some RBI documents and was able to process high volume data within seconds. “We can process thousands of documents within a Central Processing Unit (CPU)-based server and this can be scaled to million pages with a Graphics Processing Unit (GPU). We were able to prove scale on multiple fronts, handling both high volume of users and documents at the same time,” says Vishnu.
The duo was soon joined by their third co-founder and Vishnu’s long-time friend Sarath Chandra, who serves as the Product Head at Subtl.ai. An engineer from Jawaharlal Nehru Technological University (JNTU), Hyderabad, Sarath has previously worked with financial institutions and had joined the startup during the prototype stage.
Finding a needle in a haystack
Subtl.ai uses Artificial Intelligence (AI) and Machine Learning (ML) to drive its engine, with Natural language processing (NLP) being at the forefront. It combines various other technologies like cognitive modelling, deep tech, semantic scoring models, natural language processing solution, document comprehension models, relevance engine, and more to help users find the information they need.
“The idea is to make it extremely easy to create models that can directly learn from user document data.” The platform enables the user to decide which documents get added to the search, and offers full capabilities to add or remove as per their choice.
The startup’s cognitive AI product reads documents, and effectively routes any queries users have to information that is normally contained within these documents.
“You do not have to be an expert or type out the exact word. AI will understand and pull out the information.”
Expanding the use case
As the startup is still in its early stages, it is looking at multiple industries where its platform can be leveraged through the self-service Software-as-a-Service (SaaS) model.
“We are getting some interesting inbounds and are trying to keep our product as flexible as possible for its use case across sectors like legal, manufacturing and education. Having said that, our core strategy will revolve around BFSI (Banking, Financial Services and Insurance) for now,” says Vishnu.
The startup has recently received POs from Hyderabad-based Bharat Dynamic Ltd, manufacturers of ammunition and missile systems, and Kolkata-based ship building public sector undertaking (PSU)—Garden Reach Shipbuilders & Engineers Ltd.
The former would use the platform to enable its missile repair and maintenance process outside the factory, especially at border areas. “It becomes difficult even for an expert to undertake repair work outside the factory as the manuals are extremely comprehensive. Our platform would help them immediately access relevant information anywhere,” says Vishnu.
On the other hand, the latter would leverage Subtl.ai’s tech to create blueprints for ship building architecture.
The startup faces direct completion from US-based startup Glean, which was founded by a seasoned team of former Google search and Facebook engineers. Glean’s software analyses the wording of a search query and shows all the results that correspond to it, whether they are located in Salesforce, Slack or another of the many applications that a company uses. Houston-based startup ThoughtTrace also offers similar technology but is restricted towards legal contracts.
There are other big players in the market like Microsoft’s Azure Cognitive Services and Amazon’s Kendra that offer similar value. “These search tools are smarter than a ‘keyword search’ but are still not fully semantic. There is also a big set up cost involved for these services,” the founder says.
Subtl.ai plans to follow a subscription-based revenue model and charge a standard annual fee of Rs 600-700 per user.
These are early days for Subtl.ai, although it has boarded three clients to build its minimum viable product (MVP). It is currently focussed on piloting its platform with SBI and gaining a strong foothold in the market.
“We plan to showcase the use case of our technology for multiple industries to drive its adoption. We are seeing some opportunities in the US and European market across financial institutions. We see a wide variety of applications across customer service support, sales, lending support, digital transactions, drug manufacturing and so on,” says Vishnu.
Given the overflow of digital data, Subtl.ai’s tech platform saves a lot of effort, and makes many tasks more productive by taking over the mundane and routine tasks. The use case is unlimited as industries continue to streamline their process and have a quick turnaround of critical information from the endless pool of information.
According to a report by Research and Markets, the global market for Knowledge Management estimated at $381.5 Billion in the year 2020, is projected to reach a revised size of $1.1 trillion by 2026, growing at a compound annual growth rate (CAGR) of 19.8 percent over the analysis period.