Tanul Singh, an AI researcher at LevelAI, is a Global Brand Ambassador at JarvisLabs.ai. He had his light bulb moment when he realised mechanical engineering was not his thing. Later, he discovered Kaggle and started actively pursuing NLP. Tanul put his back into it and became a Kaggle Grandmaster in 8 months. 

In an interview with Analytics India Magazine, Tanul has shared his journey to becoming a Kaggle Grandmaster and his passion for NLP.

Excerpts:

AIM: When did you discover your love for data?

Tanul: I always had an interest in computer science, from my first exposure to programming in 10th grade. But due to a sub-par rank, I ended up taking mechanical engineering as a major in my college. In my third year, I realised mechanical engineering was not my thing, and I always wanted to study coding. I had read an article about computers being able to detect digits automatically, and I found it cool. So I decided to explore the area. I didn’t even know the term Data Science or ML then. I reached out to my school senior, a CSE graduate from NIT Hamirpur, and he introduced me to Andrew Ng’s course. I was juggling between mechanical and ML stuff for the rest of my third year. After I got placed with a core mechanical company in my final year, I thought it was time to focus on my passion seriously. Finally, I started learning Python in December 2019 and haven’t looked back since.

My professional journey has been like a dream so far. It wasn’t a smooth ride in the beginning. Being a mechanical engineer, I used to face a lot of issues. At the start, there was a lot of self-doubt (whether it is the right decision, will I be able to compete with people from CSE background, etc). Not a lot of people believed in me. But I pushed myself because I knew this was something I loved, and even if I failed, I would not regret it.

AIM: What is it about NLP that excites you the most?

Tanul: NLP is one of the verticals of ML. In the beginning, I was really scared. So I decided to start with an NLP competition in Kaggle to get rid of my fear. I worked hard in my first competition: I wrote an EDA kernel, and people loved it. I learned from the code shared in the forums, reached out to people to clear doubts, and eventually secured the 44th position with a silver medal. That changed my life forever.

NLP is very challenging as it deals with human lingo and human emotions, which is very dynamic. However, it’s very intuitive at the same time. You can feel what the machine is trying to learn and how it’s doing that. It’s very similar to how a child learns to read. This is what excites me the most.

AIM: What makes a good machine learning engineer?

Tanul: For me, a good ML engineer has significant knowledge of the latest development in the field and has a good grasp of the basics. She knows when to use an ML model to solve a problem. I have seen people use ML models for problems that don’t need them. A good ML engineer can frame an ML problem from a business problem and deliver a solution within the specified latency. After all, if your solution cannot add value, it’s not a good solution.

AIM: What made you interested in Kaggle? 

Tanul: Two specific things got me interested in Kaggle:

1) Kaggle is not just a competitive data science platform but also a big community of the world’s best data scientists. You get to interact with them, learn from them, learn from how they think and from their solutions, etc. I think Kaggle is the best platform to learn new things and get mentorship.

2) Coming from a mechanical background and a college with no exposure to AI/ML, I knew I had to make my name in the community to shift the domain and land a job in data science. To that end, I cannot think of a better platform than Kaggle.

I am a person who loves ML as it is very challenging. The most exciting thing about Kaggle is the variety of problems that I get to solve, both on difficulty and domain levels. Kaggle has given me a lot, and my dream is to become a 4X GM one day.

AIM: Can you take us through your journey to become a Kaggle Grandmaster?

Tanul: I felt content and happy: A mission I had set out for became a reality in eight months. The more competitions I participated in, the better I got. While I am satisfied with what I have achieved, I believe there is still a long way to go, and I want to become a competitive GM. Looking back, the only challenge I had was the fear of Kaggle as it was so overwhelming, and I didn’t know where to start. But as soon as I got over it and started doing what I knew and learned what I didn’t know, it was just smooth sailing.

AIM: What’s your favourite Kaggle competition?

Tanul: I have done a lot of interesting competitions, but my favourite will always be my first competition: Tweet Sentiment Extraction. I started the competition with zero knowledge of NLP and went on to win a silver medal. My approach was very simple: do what you know, learn what you don’t. I started with what I knew–EDA and understanding of data–and learned what I didn’t– transformers, LSTM, GRU’s, attention, etc. Abhishek Thakur has been an integral part of my journey, and I have learned a lot from him. In this competition, I learned everything from his videos and kernels. I read each kernel and discussion thread published right from day 1.

AIM: What are your thoughts on the future of data science?

Tanul: Data science is an ever-going field. Every day you have a new paper, a new technique. It’s sometimes overwhelming due to the sheer pace of things. As for the future, I think the max potential is yet to be achieved, and no one can fathom what it would look like. So we can just wait, enjoy what we have and learn as we go along.