Coseer: The Cognitive Computing Accurists

Praful Krishna, CEO In the era of automation, labor-intensive manual information mining approach employed on lengthy texts across industries continues to be the bane of many professionals. While this approach is critical to any language-driven workflow, there is hardly any other way out; as a result the workflows haven’t been digitized. Cognitive technology, with its ability to process natural language, can automate these repetitive and predictable workflows and free-up the employees’ time to concentrate on tasks that are productive. However, even the bellwethers in the industry are striving to deliver the bold promises of cognitive technology in the wake of high costs, and tedious, time-consuming process for training the systems. Praful Krishna—a specialist in the field of artificial intelligence and cognitive computing—cracked the code to this conundrum by laying the foundation for Coseer. The company’s platform understands the language the same way a human does—process language without grammar and develop new language comprehension on-the-fly—to automate any workflow based on language or unstructured data. It helps automating repetitive workflows and trains for each customer’s application within a span of 4-12 weeks with 95-98 percent accuracy.

Since its outset, the firm is endeavoring to build a future where humans focus on creativity and judgment, while cognitive technology deals with everything mundane and routine. Given that humans are engaged in the domain of ideas, Coseer converts nebulous, non-homogeneous and unstructured language inputs to ideas away from keywords. In essence, Coseer ingests written or unstructured data, either from the internet, from consumers or from data bases, and performs specific tasks or workflows automatically. With AI entering the mainstream, Coseer is already a step ahead by ushering in tactical cognitive computing in humdrum routine work. This is where the firm’s technology is setting a benchmark by emulating the complex human thought process. In a nutshell, Coseer is making the transition, from deterministic, programmable systems to probabilistic, cognitive systems, a reality.

The Dawn of Cognitive Computing

To begin with, Coseer’s team spots out repetitive language driven workflows involving monotonous manual steps. This is followed by emulating the workflow on the Coseer platform and training the systems with relevant customization to reach utmost accuracy.
Powered by advanced AI and NLP sciences, the firm’s artificial intelligent platform understands and readily learns new things, even new languages, to deliver cost-effective automaton of workflows for mundane decision-making.

Coseer’s tactical cognitive computing approach builds highly accurate solutions that train uniquely for each problem within weeks, and at a fraction of cost of its competitors like IBM Watson

Coseer comprises of a suite of tactical cognitive computing software—a continuum of technical capabilities that train for clients’ applications. DocSeer, a Coseer product, specializes in reading documents followed by building a cognitive model that ingests the ideas in the document and interfaces with a chatbot, natural language search, personalized interfaces, or more; to find an answer to a given problem in a natural way. “For example, consider the words ‘red ball’ and ‘planet Mars’. These are two different ideas but there is an overlap between them which a human can intuitively judge but a simple machine cannot. This is where DocSeer is bringing about a change by managing and figuring out the similarities. DocSeer’s ability to associate objects in the idea context is what gives us a leg-up in the industry,” extols Krishna.

Built on DocSeer is LongSeer—a summarizer for millions of documents, enlisting the customized insights into bullet points, a scorecard, or a two-page summary, as per client’s needs. In addition to this, the firm’s InfoSeer assembles all of the information in a structured format enabling big data technologies to take over. RedSeer allows redacting and preventing all the sensitive information contained in documents from leaving enterprise firewall. In a dexterous manner, RedSeer ingests, understands and eventually redacts the sensitive information in a way that the initial and relevant sense of the structure or strategy continues to be clear.

With a dramatically different approach from IBM's supercomputer—Watson— Coseer’s product suite is steering ahead with cost structures that are a tenth of Watson’s costs, and are accessible to individual teams with limited budgets. Needless to say, Coseer’s products are carving out their niche to clients from diverse verticals like Finance, Healthcare, Manufacturing, and Legal. For instance, Coseer worked with a healthcare vendor that provides clinical decision support to hospitals, doctor’s offices and other providers.
They have built expert systems based on legacy technologies, however the biggest bottleneck for the process is to manually capture actions, questions, diagnoses, and other structured information from documents that can be 800 pages long. Coseer recognized that this process is completely automable. “We designed a solution that can annotate these texts into a structured database within milliseconds, saving weeks’ worth of manual effort. The solution provides for machine assisted verification as well, when necessary,” explains Krishna.

100 Percent Accuracy is just a Hypothesis

Very high accuracy is the hallmark of tactical cognitive systems. “Consumer applications can afford to be relatively more inaccurate because the tasks are usually not mission critical, and there is instantaneous human feedback. For enterprises, anything below 95-98 percent accuracy does not work,” stresses Krishna. It is near impossible to get general purpose AI to reach that level of accuracy at current state of art. Coseer’s tactical cognitive computing approach builds highly accurate solutions that train uniquely for each problem within weeks, and at a fraction of cost of its competitors like IBM Watson. “Although IBM Watson defeated humans in the TV show ‘Jeopardy’, it didn’t answer all the questions accurately,” points out Krishna. “Coseer is a platform that trains over specific data for a particular problem of a particular client specifically, for 2-3 months. This is how we achieve such high level of accuracy.” Such degree of accuracy also evolves from Coseer’s unique capability of constantly learning new things and experiences, and receiving feedbacks on what is not accurate.

Paving Way for Emulating Human- Like Capabilities

For Harvard Business School, IIT and McKinsey alum—Krishna, it was least surprising to become a Silicon Valley hipster launching a startup of his own. However, he embarked on an entrepreneurial journey at a time when cognitive computing had just started gaining steam with people promising the moon yet didn’t get even close. Setting the record straight, Krishna states, “Ours is not a system that argues and reads in a way like Shakespeare but is a very tactical and very purpose-built solution that can automate language-driven workflows.” In days to come, the firm has plans to foray deeper into the industry by building multi-language capabilities. In addition to this, Coseer’s upcoming project—a human brain like central repository, live and available for all Coseer apps—looks promising. “A big part of the human intelligence is the experiences they accumulate. Our applications are learning these experiences in isolation across the board. With this central repository we can aggregate those experiences and ideas learnt followed by adding intelligence through internet, to help the applications sort out any confusions and issues,” says Krishna. Sure enough, the CEO’s zest to create ‘smarter’ world and pursue greater efficiency and accuracy holds a promising future for Coseer.


San Francisco, CA

Praful Krishna, CEO

Coseer builds cognitive computing technology to automate routine enterprise decisions for language based workflows