Analytics India Magazine got in touch with Debdoot Mukherjee, VP of Artificial Intelligence at ShareChat, for our weekly column My Journey In Data Science. Debdoot has over 12 plus years of experience in machine learning related projects while working for organisations like IBM Research, Myntra, Hike, and ShareChat.
Debdoot, after completing his bachelors from the Institute of Engineering & Management, Kolkata, pursued M.Tech from IIT Delhi in computer science and engineering. After post graduation in 2008, Debdoot joined IBM Research as a software engineering researcher. He used to work on software productivity tools, which involved information extraction and semantic search. “In our research group, we applied machine learning techniques to improve the productivity of knowledge workers,” said Debdoot. “My first interaction with machine learning was at IIT Delhi, where I learnt the fundamentals of machine learning. However, most of my learning happened while working for various companies,” he added.
Although Debdoot came across machine learning during his M.Tech, he got interested in the domain while working for IBM Research. He was hired as a software engineering researcher, but machine learning was becoming the go-to technology for almost every solution he used to develop, which got him hooked in the domain.
During his six years at IBM Research, Debdoot had to learn new machine learning techniques as the domain was picking up the pace. He used to read books for learning various methodologies, which could be used to solve problems in this work. Unlike today, where aspirants have a wide range of courses to learn from, Debdoot did not have the opportunity to learn through various MOOCs. He used to stumble upon a problem, and then pushed himself to find the tools and algorithms that would help him in mitigating the challenges. “Given a project, I went back to first principles and used to find the right approach, techniques, and tools for my problems. This allowed me to have focused learning rather than learning everything that the domain offered,” explains Debdoot.
Debdoot suggests that aspirants should learn the foundation and machine learning basics from online courses or textbooks. But, picking tools and techniques should be based on the kind of problems one is trying to solve. Problem first approach helps aspirants in avoiding the trap of buzzwords in data science. “Today, learners are invested in Kaggle to hone their skills, but nothing matches the problem first approach,” says Debdoot. “Competitions hosted on Kaggle are bounded by the scope and formulation is clear from the statement itself. This does not develop your critical thinking skills, which is essential for succeeding in data science. When you solve business problems, you evaluate a lot of algorithms and techniques before implementing them. Open-ended projects do not necessarily have clear solutions, thereby enhances the data intuition in aspirants,” he added.
Although Debdoot had the flexibility to work on almost any problem at IBM Research, he decided to move on to Myntra to work on large scale real-world datasets. At Myntra, Debdoot, along with his team, developed various hyper-personalised solutions for fashion e-commerce, which he said was very early in the data science market. After a little more than one year and six months, Debdoot joined Hike in 2015 as Director of data science. At Hike, he worked on numerous projects that included NLP, social network analysis, and computer vision. As Hike is a social networking platform, it brought in a diverse set of challenges, particularly in interpreting Indic languages. The team at Hike also published numerous research papers at top international research conferences. And today, at ShareChat, he is involved in delivering the right content at the right time for users to enhance engagement on the platform. What fascinates Debdoot is that ShareChat can impact a huge Indian audience. Due to the state of the art recommendation systems of ShareChat, the platform renders relevant content to users without even having to follow the content creators. This is done by applying AI to achieve a deep understanding of the users on the platform.
As a part of his job, Debdoot also hires data scientists and seeks three competencies in applicants. The first one is their grasps and practical knowledge of statistics, machine learning, and deep learning. Secondly, he looks at software skills like working with distributed systems, large scale data and more. Finally, the evaluation is done on their product thinking — the ability to slice data and experiment in the right way. However, Debdoot said that one cannot be good at all of these skills. Thus, one who is above average in any two competencies would be a potential hire.
Advice To Aspirants
As a piece of advice to aspirants, Debdoot said beginners should not fall for buzzwords as new techniques will keep evolving in the ever-changing data science landscape. The techniques that one uses today will be different from the methods utilised in future. “Learning a technique and completing an assignment on online courses will not help if aspirants do not apply them to real-world problems. Consequently, they should go backwards — find a problem and then learn the necessary techniques to solve the challenges effectively,” concludes Debdoot.