Interview With Karthik Kumar, Director of Data Science at Epsilon

For this week’s practitioner’s sequence, Analytics India Magazine received in contact with Karthik Kumar, Director of Data Science for Auto Practice at Epsilon. He leads the info science capabilities constitution the place they develop superior analytic options, merchandise, methods centered on the acquisition, engagement, and retention of the shoppers for the highest Automotive OEMs. 

AIM: Tell us a bit about your academic background

Karthik: I acquired my Master’s in Management Information Systems from University at Buffalo, SUNY. Before that, my undergraduate diploma was in Science. My ardour for knowledge science helped me study from prime universities akin to MIT, Stanford, and Indian Statistical Institute. These beneficial academic experiences, programs, and an urge for steady studying and curiosity helped me construct my profession in knowledge science.  

AIM: How did your knowledge science journey start?

Karthik: Data science has at all times fascinated me proper from my tutorial years.  What has at all times intrigued me is how knowledge could be transformed into beneficial insights with a magic sauce algorithm. At an early stage in my profession with a number one consulting and tech agency, I began to use these learnings on actual use circumstances. 

One of my preliminary tasks was the place I used to be requested to resolve a human capital problem to proactively determine excessive performers who may attrite which was a key focus space for enterprise. The studies and dashboards which we had developed had been solely a reactive strategy and therefore the main target was to develop a proactive resolution that might be an early warning device to take needed motion to de-risk attrition with excessive performers. This was step one in the direction of creating a machine studying mannequin which was created utilizing a sophisticated algorithm the place it predicted the likelihood of a set of excessive performers anticipated to attrite within the subsequent “x” variety of days. This acted as an early warning device for the enterprise to take needed actions. 



This resolution was additionally offered at Indian Statistical Institute throughout an Analytics summit the place it received the “Best Predictive Analytics Paper” award. This was a catalyst undertaking which propelled my Data Science journey the place I began creating ML options for Retail, Human Capital, Telco, Finance and Media shoppers. 

AIM: What had been the preliminary challenges and the way did you deal with them?

Karthik: The preliminary challenges had been to get an understanding of the trade and area. For instance, after I was roped in for an trade undertaking for a number one international mining group that focuses on discovering, mining and processing the Earth’s mineral assets, we had been supposed to construct a proof of idea to determine anomalies of the ores after the crushing course of; for which I needed to spend intensive time to know the mining course of to obviously outline the goal variable to construct the machine studying resolution. 

The subsequent problem was to work throughout numerous groups to assemble the precise knowledge, for which, we needed to translate quite a lot of the folks’s data right into a course of after which flip it into high quality knowledge. With the excessive mannequin growth time, the actual problem was to develop a close to actual time resolution with steady knowledge feed. This is the place we roped within the engineering group to assist us with a knowledge structure resolution to feed enter knowledge to the machine studying mannequin to attain and determine the anomalies. 

Karthik Kumar along with his group @ Epsilon. (Credits: Karthik Kumar)

AIM:  What does a typical day appear like for you at Epsilon?

Karthik: Each day the endeavor is to get higher by not less than 1% with innovation, stakeholder engagements, execution of the info science tasks which result in excessive enterprise influence, and buyer satisfaction. I join with my group to ask them the precise questions, present them the required steerage and help to construct the new-age knowledge science options and merchandise. The gifted group of knowledge scientists, engineers and analysts in flip assist me study repeatedly.

To guarantee we’re abreast of the most recent developments within the space of machine studying and AI, we’ve a robust group engagement tradition the place we meet each month to deep dive on key subjects,   focus on intimately a analysis paper, or have colleagues share their greatest practices with the group. These knowledge-sharing classes assist us collaborate, share concepts, and particularly throughout these testing instances of the pandemic, this helps to maintain our modern, inventive minds challenged even in a digital, hybrid office!

AIM: How do you strategy a knowledge science drawback?

Karthik: As they are saying, “Data is the new code”. The machine studying code is simply a small portion of the puzzle and wouldn’t suffice to take the mannequin from a POC stage to manufacturing. Deployment is a course of the place it’s a steady knowledge movement and studying journey, making ML an iterative course of. Hence sustaining top quality in all phases of the ML life cycle is a very powerful process. 

The first step is to know the enterprise drawback and to translate it right into a statistical/machine studying drawback. In this expedition, the standard of the info is important and that is the place a knowledge scientist has to spend most of his efforts to raised comprehend, and remodel the info to know its traits to construct a strong machine resolution resulting in profitable enterprise outcomes. 

The quantity of labor on mining the precise knowledge, bettering and understanding the info is a very powerful step which I’d emphasise on my tasks. An intensive function engineering from the info would assist construct a robust knowledge science mannequin versus iterating the fashions on a set knowledge set. My tip to budding knowledge scientists can be to speculate most time in gathering the precise knowledge, exploring and creating the options innovatively.

“At Epsilon, we are tool-agnostic where our engagement models are flexible to ensure it is aligned with our customer requirements, infrastructure, use cases.”

Karthik Kumar

80% of knowledge science work is about knowledge which is the meals for AI, the high-quality knowledge with the precise preparations can be a very powerful step within the AI programs. Though it appears much less fancy to discover options, create the multilevel variate plot, remodel knowledge and create knowledge options, these are the constructing blocks for a stable mannequin. 

AIM: What does your machine studying toolkit appear like?

Karthik: At Epsilon, we’re tool-agnostic the place our engagement fashions are versatile to make sure it’s aligned with our buyer necessities, infrastructure, use circumstances and so on. Our Offerings are actionable instruments facilitating enhancements in all advertising combine components throughout the client journey to drive acquisition, engagement, and retention.  We develop our superior analytics choices, merchandise, providers and, machine studying capabilities with a large spectrum of analytics instruments, languages, platforms akin to R, Python, SAS, Spark, PySpark, H2O.ai, Databricks, Dataiku, AWS and, Azure cloud ML platforms, AWS SageMaker. With our Data science COE neighborhood, we uncover DS purposes, instruments, strategies to repeatedly evolve with the brand new AI/ML developments. 

AIM: There is quite a lot of hype round AI and ML. Which area of AI, do you assume, will come out on prime within the subsequent 10 years?

Karthik: The hype is actual. I wish to borrow a few quotes right here:

“AI is the new electricity” – Andrew Ng.

“AI is probably the most important thing humanity has ever worked on. I think of it as something more profound than electricity or fire”

See Also


– Sundar Pichai

There are quite a lot of runaway horses, however the promise is actual if you’re methodical concerning the knowledge you have got and the use circumstances you clear up for; there’s a pot of gold on the finish of the rainbow for certain.

A McKinsey study estimates a 13 trillion US {dollars}’ value of US financial progress and worth creation as a consequence of AI. Though at current, the AI funding is rising quick, it’s dominated by digital giants akin to Google and Baidu. The thrilling a part of AI will probably be its growth and adoption throughout industries and we’re already seeing that on an enormous scale within the automotive sectors. 

Today 99% of the AI financial worth creation is thru one sort of method known as supervised studying which is an enter to output relationship. An automotive-related instance is a know-how utilized in autonomous vehicles, the place an enter is an image of what’s in entrance of the automotive and the choice as an output feeds into automotive programs. With the precise business use circumstances, within the subsequent decade, AI goes to remodel industries akin to  Retail, Media, Education, Healthcare, and journey, which at current has low to medium adoption of AI.

AIM: What does it take to land a machine studying job at Epsilon?

Karthik: At Epsilon, we put our folks on the heart of what we do and supply them with difficult alternatives the place they get to work with a few of our world-class knowledge belongings, applied sciences, and platforms throughout industries. We even have vibrant analytics and knowledge science practitioner neighborhood internally, that works carefully with prime expertise to study, collaborate, and get impressed by the analytics work throughout the varied follow areas by analytics occasions akin to hackathons, roadshows that we host usually. We lengthen these ‘Hackathon’ applications externally to determine prime expertise folks from round India, APAC marketplaces and in addition we’ve a yearly Campus graduate and internship applications the place we work with the highest faculties to rent freshers to our Epsilon household who add to the required creativity and, new approaches to resolve the advanced knowledge science use circumstances.  

Any aspiring candidates particular to ML roles needs to be curious, show a excessive stage of ardour and showcase examples the place they’ve solved machine studying use circumstances. Machine studying is an evolving subject and the candidate ought to at all times be on a steady studying path and would suggest showcasing their contributions on platforms akin to Kaggle, Github, and LinkedIn.

AIM: What books and different assets have you ever utilized in your journey? Few phrases for the aspirants as effectively.

Karthik: It is vital for aspiring Data Scientist to develop a robust basis in statistics, I’d extremely suggest Introduction to Statistical Learning, a guide that might assist construct on the fundamentals of statistics data and when you graduate, at an intermediate/superior stage, the subsequent really useful guide can be Elements of Statistical Learning that cowl superior subjects like neural networks, matrix factorization. 

Also,  one must construct a stronghold on not less than one programming language and with the info dealing with capabilities, parallel computations, giant integration capabilities, and communities, I’d suggest Python open-source device which has probably the most superior libraries to resolve the new-age knowledge science issues. The Applied Data Science with Python Specialization by Coursera helped me acquire good utilized data of Python and to get hands-on with toolkits akin to pandas, matplotlib, scikit-learn, nltk. These are only a few to call which have helped me in my Data Science journey.

As a newbie, one may be overwhelmed with the varied choices obtainable to develop the Data science abilities in programming, arithmetic and statistics. Be affected person, stick to at least one knowledge science programming language, develop a dense understanding of mathematical and statistical abilities. This will assist freshmen construct a robust basis for his or her Data Science careers. And lastly, getting hands-on expertise may be very essential that will help you apply the theoretical to the sensible points to witness tangible outcomes. I’d recommend new knowledge scientists take part in competitions on platforms like Kaggle which is able to assist in the sensible software of data and to study from the robust knowledge science neighborhood. 


Subscribe to our Newsletter

Get the most recent updates and related affords by sharing your e-mail.


Join Our Telegram Group. Be a part of an attractive on-line neighborhood. Join Here.

LEAVE A REPLY

Please enter your comment!
Please enter your name here