NLP is a very broad term and industry has been using it everywhere. – Shivam Kotwalia

Interview with Shivam Kotwalia

Shivam Kotwalia

Lead Data Scientist
Delhi, India

Q1. Please share your educational and professional journey with us?

I Completed my B.Tech with specialization in Electronics & Communications, joined industry as an ERP consultant helping clients convert their complex ideas into business. My life was all breeze but until one day I participated in a hackathon, poured every sweat into and still achieved consolation prize 🙁 I contacted the I & II winner and they apparently used something called Machine Learning. This caught my attention and I started consuming ML, after a while I shifted to ML CoE and established the first client facing ML team. Later I joined a Management Consulting firm where I held 2 roles, first helping clients achieve impact through ML and second codifying and generalizing these ML systems into solutions.

Q2. You did Engineering in Electronics then what did attract you towards Data Science and Machine Learning Domain?

The good thing about Data Science and Machine Learning is that it can be applied in any domain, I recently completed a solution with Electronics domain – where traditionally doing a cost breakdown of any electronic PCB (printed circuit board) would take 6-7 days depending on the component mounted on it, but with the help of computer vision we can count/mark/edit/ calculate cost any numbers of components on a PCB within hours (including QC).

So, yes I can clearly say it has made an easy way for students from other branches to enter the market and disrupt it with ML/DS.

Q3. Is NLP part of Deep Learning or something different? Where can anyone use NLP?

NLP is a very broad term and industry has been using it everywhere.

In my understating text is a cognitive vertical like images, speech, the relationship that exists between text and NLP is similar to images and image processing. Deep Learning has entered the domain quite lately and revolutionized it, with the release of attention based models, transformers – NLP is going wild.

Regarding, the usage – text is more freely available than any other form of data. You just name the domain and NLP can be applied, from building a fail safe campaign for M&S team or check your employee NLP score for HR or product categorization & supplier negotiation for Procurement teams.

Q4. Which one thing do you want to change in Data Science Domain and why?

I want to disrupt the way ML is organized in firms, people have dedicated ML teams who will build world class models but will fail to deploy them either due to lack of infrastructure (MLOps) or lack of a structure (Agile in ML) or disjoint functioning of DS team, DE team, Dev team and Infra team.

I have seen my clients applying the same Agile routine in ML like they do in their any other traditional software team, which is not correct. You can’t bind your ML team to a sprint and expect a “95% accuracy model” without giving them the proper data. Or, what is the use of this super awesome model, if it hasn’t been deployed scalably  or not monitored for leakage or drift.  

Q5. Which one is the hardest working part of any Data Science Project?

Both, pre and post operations are quite interesting and hardest to work. A good model can’t be made without good data & a model can’t give good results until and unless monitored correctly.

Q6. How do you see Online Courses of Data Science? Do you recommend any online course for Data Science?

I believe today for learning Data Science you have a plethora of resources, based on your current skill set, based on your current domain – internet is filled with them. You just need to Google 🙂

I don’t recommend any specific course but I believe in T-based learning, play on your forte and revolve around it. Second, I strongly advocate for “learn while you work” strategy. I have seen people not touching DS problems because they haven’t completed the Maths or Statistics or something else, that day is never going to come.

I see currently an issue that – how does one keep themself updated with the everyday changes/updates/papers/library  in DS/ML. One needs to  be very cautious which data to be consumed and how much to consume.

Q7. Which one thing do you want to change in yourself and why?

I have somehow learned them, but want to develop more

  1. how to channelize your thoughts while solving a business problem
  2. How to present/showcase your finding of ML and make them lay man understandable.