For Data Science Extraordinaire Karun Singla , Applied Machine Learning is about transforming theory to tangible business impact. With a passion for technology and a relentless drive to solve complex problems, Karun has carved a niche for himself in the world of Artificial Intelligence. With his years of experience across global big tech companies - leagues of paytm, Alibaba, Dailyhunt and Revolut, he quotes “We are in the era of Big Data where Petabytes of data is generated everyday through financial transactions. It is more crucial than ever to extract insights to gain a competitive edge, bolster innovation and improve User experience”.
Karun hails from India, and holds a Bachelor's Degree in Technology from prestigious Delhi College of Engineering. Right from his engineering years, he witnessed the inability in large organizations to harness data to solve their own problems. After demonstrating significant process improvements with Machine Learning at India’s largest Conglomerate Aditya Birla Group, Karun made strides in the fintech wave with AI. He spearheaded the analytics in leading fintech companies across India and Nigeria. Over the years, he has widened his expertise in Machine Learning across domains of healthcare, Ecommerce, Adtech and to rise to one of the most promising ML talents in India.
He is currently driving growth initiatives for UK based fintech - Revolut. To transform Revolut into the world's first truly global bank, Karun believes data science holds the potential to unlock more business opportunities.
Going Above and Beyond
Outlook India visited Sherlock Institute of Forensic Sciences , Delhi headquarters. Karun has taken upon himself to address the gap between Industry and Academia. “Karun is our prominent industry partner in AI. He is highly respected in the SIFS community for his way of making AI/ML intuitive for our students.” quotes Dr. Ranjeet Singh, CEO of SIFS. We interviewed Karun to understand his vision better:
Q: What are the most important learnings of your career?
A: I believe Data Science is interdisciplinary. The biggest learning is how to apply skills acquired from one domain to another. I used learning from modeling low-event rate health conditions at UnitedHealth Group(global healthcare insurance leader) to improve Fraud detection at Revolut. Both of the problems involve maximizing recall and precision. At one instance you predict patients likely to encounter Alzheimer's, at another you try to catch a fraud transaction. Both problems are like finding a needle in a haystack. The same can be said about recommender systems in Advertisements- Optimising Click thru Rates during my stint at Dailyhunt is another example.
Q: Is there any difference between a Data Science researcher and a Data Scientist in a product led company?
A: In an academic setting professionals need to understand AI is not just about building models. A highly accurate model can even have a negative impact on the business metric. Revisiting fraud detection here, you can build an engine that has very high accuracy, but what if it only predicts non-fraud transactions correctly, or has poor precision. This is where Data Scientists need to think deeper and try to formulate the modeling solution keeping in mind what is the business metric I am trying to improve.
Q: What are the other skills, apart from knowledge of ML/AI algorithms that are important for emerging data scientists?
A: Engineering! The data science professionals need to understand that the model is merely 20% of the job. Real data science extends beyond jupyter notebooks. They need to have a deeper understanding of engineering concepts around deployment, CI/CD pipelines, model monitoring. It's important to control spurious behavior of the models in production. This is also another key difference between data scientists in a real-world vs academic setting.
Q: What are your biggest challenges at work? Can you give some real life examples?
A: Production systems or deploying ML services are not short-term projects and involve multiple stakeholders. The biggest challenge I have faced is how to convey algorithm design to non-engineers, say product managers or sales heads. When I was building a recommendation system for Paytm using a clustering based approach and I wanted to convey relevance of each of the cluster, i went to my VP and said - You want to buy a pair of jeans, I give you 3 options to choose from-
- First shop has all the brands but the pricing is not competitive.
- Second shop has the most competitive pricing but their variety is meh and the products could be fake.
- Third shop has 90% of the most popular brands and their pricing is deviating 10% from best sellers but they are fully authentic.
These 3 shops represent statistics of 3 clusters and now don’t you think my solution is intuitive?
Q: How do you segregate Good data scientists from outstanding data scientists?
A: A good data scientist solves the given problem. An excellent data scientist identifies the problem and solves it. This is where product understanding comes into play. Until and unless you don’t know the business like the back of your hand you cannot identify the pain points of the business.
Q: What impact are you trying to make with SIFS ?
A: I believe AI is ever evolving. Mentorship is a two way street. My mission is to bridge the gap between Industry and Academia. It is equally beneficial for me, because it helps me to stay inline with the latest advancements in technology and of course I learn equally from my mentees.
He concludes the discussion by quoting Albert Einstein - “ The more I learn, the more I realize I don’t know”