SUMMARY
CogniTensor started predictions for energy and aluminium to help companies better manage and procure their raw materials
AI is not just about the product but also about having the data scientists to come up with results that the customer wants, says cofounder Ashish Airon
He talks about the next wave of AI, data security, talent gap in AI and more
The move from paper to technology has changed the business landscape in India drastically over time. Organisations now have a lot of data with them thereby enabling decision making to be highly data-driven. Just having the right data is not enough, what gives one an edge over competitors is how best one can utilise the available data to derive insights that help in making crucial business decisions.
“The AI segment has grown since the year 2012 as there are a lot of opportunities with new models in the market. However, many models have not yet been applied to the industry due to the hockey stick graph,” said Ashish Airon, cofounder of deeptech startup CogniTensor.
Established in April 2018, CogniTensor essentially collates data from various sources, processes it and provides actionable insights to businesses across sectors. The insights help companies to make the right decisions on the procurement of raw materials and other aspects, thereby saving costs and turnaround time.
Cofounder Pankaj Mathur said that CogniTensor works in two ways. “In areas such as the energy market and aluminium, where we have gained domain expertise over a period of time due to various successes, we offer our insight and analysis directly to customers,” he said.
For other sectors and segments, CogniTensor gets into partnerships with companies. Using their data, the AI company generates customised reports. “Like we did with Altran, an engineering services company now acquired by Capgemini,” added Mathur.
NITI Aayog in its “National Strategy for Artificial Intelligence,” stated that India’s capabilities in AI research are rather limited, both in quantity and especially in quality (disappointing impact of research produced). It further said that the research community is rather confined to a handful of academic institutes, and relies on individual brilliance rather than institutional competence.
CogniTensor started operations in the UK and Middle East before entering the Indian shores. Airon believes that the skillset in India is very different from the one in other countries. “AI is not only about the product but also about having the data scientists, the right people to come up with the result that the customer wants. In India there are many job openings for development operations that revolve around managing the structure for AI, deploying the code or maintaining the pipeline but universities in other countries focus heavily on the research part. Thereby the gap is huge,” he said.
Due to the lack of the right educational courses in mainstream universities, the AI and deeptech industry in India struggles to find the right talent. Seconding this, Airon said that the desperation is really high to get into artificial intelligence as the salaries are good and the demand is high but the talent is just not there.
Both founders bring a wealth of knowledge to the table. Airon worked as a researcher in Germany in natural language processing and holds a post-grad degree in computer science from Oxford University, where he specialised in deep learning. While Mathur has experience in creating and scaling businesses for top MNCs, Arun Aggarwal had a background in the IT and telecom sectors.
“Our two core strengths lie in the research and the external data that which we capture,” said Airon.
CogniTensor’s product has three components. The first is an ingestion piece with the ability to handle and take in data from all different sources. The second is the brain — the real engine — which generates the insight and the third piece is where all the dashboards can be accessed through an app or website.
With the core focus on commodity pricing, CogniTensor started with predictions for energy and aluminium to help companies better manage and procure their raw materials. It plans to enter new markets in both, products and geographies.
“We are expanding in the Middle East, where there are large smelters for which we are providing price predictions for 15 days to one month in advance of aluminium and energy with an accuracy of more than 90%,” said Mathur.
He admitted that while the company was testing waters in various markets at the start, it realised the need to focus. “As we grew, we realised the importance to focus and cater to an area which is very unique, where the benefits are clearly visible and needed,” Mathur added.
With domain partners, the company plans to focus on telematics for commercial vehicles to assess possible faults to enable predictive maintenance for the engine. However, this along with healthcare, are two long-term goals for the company due to lower margins. Calling it the “cookie-cutter case” the commodity markets would be the near term focus as it helps generate faster revenues for the company.
As an artificial intelligence company, CogniTensor deals with humungous amounts of data. The security of which becomes questionable. While the company claims to not store any personal identifiable information and have the right measures in place, Airon adds, “ Security has two angles to it, one is data security and the other is the security of the algorithms which have been made because they can be reverse engineered to extract data. So security is a big issue and more work needs to happen in this space.”
A problem in the neural network is a phenomenon where a network tends to learn certain things and gives biased results. Airon explained this through an anecdote, “In the UK someone had written Oxford University with white ink on their CV. While this was not visible to the eye, the CV bypassed the computer systems in place. Another example is where if you expose or write a certain symbol on the camera screen, it will blur your face.”
In both cases, the network has memorised some patterns and gives a biased result. If a person knows how the model was made, they can bypass the program. It is important to make sure that the network is neutral and does not leak any information.
Nasscom predicts that by 2022, a startling 46% of the Indian workforce will be engaged in entirely new jobs that do not exist today or jobs that have radically changed skill sets. In the data domain as well, an independent study estimated that India will face a demand-supply gap of 200K data analytics professionals by 2020.
In the IT-BPM sector, traditional software developer roles are set to transition to roles such as computer vision engineers, robotic process automation (RPA) engineers and cloud architects, among others.
At the same time, completely new job roles such as language processing specialists and 3D modelling engineers are set to arise as the technologies are increasingly adopted and deployed. With AI, such a transition would move beyond the IT sector and affect sectors such as education, health, agriculture, finance, etc, requiring the underlying skill sets.
Airon believes, “The next wave which we will be entering now is the explainable AI space. The industry has worked a lot in experimenting but now we will be working on the explaining layers. The industry focus will now be on building networks that can establish why something is happening and whoever is able to leverage and build that knowledge will definitely have a unique spot in the market.”
