When Puneet Agarwal graduated from IIT Delhi in 2008, the term machine learning (ML) or artificial intelligence (AI) was not known to many. It was long before ML and AI became industry buzzwords. In fact, IIT Delhi, which presently boasts of developments in the field of AI and ML, didn’t have a faculty in the area.
His first brush with classical machine learning came during his 3-month internship at Yahoo where he worked on a classification problem using Support Vector Machines (SVMs).
Subscribe to our Newsletter
Join our editors every weekday evening as they steer you through the most significant news of the day, introduce you to fresh perspectives, and provide unexpected moments of joy
Intrigued by the technology, Agarwal went back to his college and convinced his professor to let him work under the guidance of the professor. This gave Agarwal an opportunity to learn more about machine learning and by the end of his graduation, he had a clear career vision and knew what he wanted to pursue further.
“Somehow I convinced one of the Professors to let me work under his guidance and I did some more work on SVMs during last year at college. That was the beginning of the journey, and frankly it happened by the change since I had no clue about ML before joining Yahoo,” Agarwal tells about his initial days.
A world class experience
Following his graduation, Agarwal moved to Germany where he joined Max Planck Institute at Tubingen and continued his work on SVMs for the next few months. The institute’s lab was so advanced in terms of its application and understanding of ML, that Agarwal was impressed by the state-of-the-art facilities it provided, “And it was during my time there that I decided to pursue ML as a full-time job,” Agarwal shares.
The internship stint and the exposure to world-class technologies only proved to be more useful as he returned to India to join one of the leading tech companies as a Senior Software Developer.
Agarwal served in the position for the next eight years, where his primary task was to improve their search engine, cognitive intelligence’s personality, query classification, among many other jobs.
A seasoned data scientist with experience in the field of Deep Learning, Web Search Relevance, Data Mining, Query Understanding, Agarwal is the man behind the company’s famous chatbots designed especially for the youngsters.
In his 12-year-long career, what Agarwal witnessed was a steady growth, as he went on to become a Senior Development Engineer to presently holding the designation of Principal Engineer Manager for the past two years. Though his career-graph has only seen a rise, Agarwal says that to keep learning is something everybody has to do without fail to excel in their jobs.
This especially holds true in the backdrop of recent retrenchment by Indian IT stalwarts, where hundreds of IT employees are being laid off due to company restructuring. “In the industry, I always worked in the field of ML, so I never came across a situation where I had to completely change my skill set. However, I feel that the field is booming and I need to constantly read up new stuff and understand the latest developments, and take advantage of those in my day to day work, that is essential in today’s time,” Agarwal points out.
His take on the industry and on the essential tools.
Essential tools for a data scientist
I like Python, but do not really believe in advocating a specific set of tools. If you are looking to get started with something, I would recommend Python and Pytorch as starting points.
What’s in his Developer kit
I use Python, I also use Microsoft Excel quite a bit. I use CNTK for deep learning experiments.
Anything you learned recently:
Learning new algorithms like BERT and many advanced techniques on top of it. These algorithms are making a lot of progress in the field of NLP.
Some Of The Challenges you faced:
Keeping up with the pace of new advancements, but not losing sight of the basics is the most important challenge I face in my day to day work.
In the last 10 years, the scale of data, computing power and algorithms have dramatically changed. In the old days, we used to give a lot of emphasis on looking at the training data, cleaning it up and getting the right set of features in place to train an ML model. In the era of deep learning, these aspects have somehow taken a backseat.
How did you overcome it?
One cannot avoid looking out for insights from data, but at the same time, it is important to not ignore all the great work that is happening in the field of deep learning, striking that balance is sometimes hard but critical in my view.
Word of advice:
Start by practising by doing coursework, that will give you a good start.
I highly recommend anyone who wants to excel in the field to work on some real challenges, there are many interesting challenges that are run on Codalab or Kaggle, pick the one that is most interesting to you and dive in, that will help you understand the real issues and give you the opportunity to apply theory to practice.
Apart from that, keep reading up on new algorithms that come up, for e.g. recently a model called BERT has made significant progress in the field of NLP, you need to learn to build on top of these generic advancements and ideas, and tune to your specific problem.
What’s In Store For The Future
In the near future, I plan to continue working in the space of search, conversations and NLP in general. I feel we have a lot of ground to cover to make AI more human-like both in terms of its behaviour and capabilities. Apart from building products at my present job, I plan to continue being part of an effort to run challenges and workshops under the umbrella of “Humanizing AI” and do my bit to strengthen the AI community, especially in India.