Your DS Story: Ranjeet Singh (Roadzen)
Your DS Story: Ranjeet Singh (Roadzen)
‘Your DS Story’ is my attempt to bridge the gap between data science professionals & data science aspirants. Here new crop of data scientists will share their experiences, struggles, achievements & their advice so that data science aspirants/enthusiasts can learn and get inspired.
1. Please tell us a bit about your background?
Hi, thanks for giving me chance to share my journey in this fascinating area of artificial intelligence and data science. My educational background has been in signal processing, which gave me chance to work on image and speech data from my college days itself.
During my masters I took formal classes in neural networks and machine learning. I started my AI journey from early 2015 when I started working on my masters thesis with Mr. Sonit Singh, currently Ph.D. scholar at Macquarie University.
I started my professional career at Samar Singla’s Jugnoo based in Chandigarh as a data scientist, where I worked on solving several problems of logistics industry from automatic ride allocation to churn prediction.
Then I moved to Mumbai, started working as a machine learning engineer at Quantiphi. There I got a chance to work on end-to-end speech recognition with a very brilliant team. It was very new but challenging experience for me due to the fact that speech is very unique data in itself and we were working in the domain of trading where no speech recognition API performs well since pretty much all of them are trained on general conversations.
Currently, I work as a data scientist at Roadzen, a multinational InsureTech startup based in Pittsburgh, U.S.
Here I’m back to my favorite area of computer vision. We are solving many real-world problems of InsureTech industry.
Along with this, I also work as a visiting researcher at multi-modal digital media analysis lab [MIDAS], IIIT-Delhi. MIDAS is a research lab with the joint affiliation of IIIT-Delhi and NUS. Here I am working on several research projects in the area of computer vision.
I am particularly interested in computer vision and Bayesian inference. I see immense opportunities in this field in our country particularly. India is undergoing a deep digital transformation that shows no signs of slow down. Average monthly data consumption of Jio alone is 10.8 GB and average voice consumption stood at 794 minutes which in turn creates a massive amount of data leading towards great opportunities for Data Scientists to explore and create business out of. After US, China, and France, the Indian government is putting efforts to make the nation stand against the global challenge. Government is building a center of excellence in AI as well, proving limitless opportunities.
2. What projects you are working these days?
Few of my projects involve automating motor insurance claims process and fraud detection, one of the main problem arising in the insurance industry.
In academics, my current work involves the application of deep learning research in surveillance, some work is in speech recognition as well.
Currently working for removing/reducing human interventions in the insurance sector.
I mainly work on image data. The main challenge that we face is that we don’t get much help from openly available data-sets, open images, Image-net due to the domain that we are working. We have to spend lot of time in getting the data labelled.
It, in turn, makes unsupervised or semi-supervised methods pretty useful for us. We do use classical computer vision methods as well which are not much data hungry. We also face challenges in getting research focused talent in this area of computer vision.
In this short amount of time, I have got chance to work in very diverse areas of artificial intelligence- from speech recognition to computer vision with some work in NLP as well. Working in AI focused organizations i.e. Quantiphi gave me the opportunity to work on things before they get mainstream. And all these experiences make me confident and let me my future moves properly and accurately.
3. How your day to day job looks like?
This is for aspiring data scientists. In the real world data science project, there is only 15–20% is data science, rest is software engineering only. When you are in college or taking MOOCs, the easiest part is getting data, you are only supposed to build machine learning models. While in Industry, the hardest part is data.
My typical day goes like:
20% — Building data pipelines
20% — Modeling
10% — Understanding business requirements
50% — General software engineering
4. How you started with DS or transitioned into DS?
As I mentioned earlier, I started this from my college itself. I got inspired by Fei Fei Li’s Ted Talk — ‘’How we teach computers to understand pictures’, which made me chose my thesis topic in computer vision. From then the journey is going on. From my early college years itself, I was inspired by Alan Turing. I also plan to do a PhD in the area of reinforcement learning in near future.
In industry, the majority of data is unstructured and dealing with unstructured data is very difficult and time consuming. Along with that when it comes to feature engineering, this is an art, and to do this in a better way — one needs to be an artist as in, you have to take domain knowledge and think from different angles.
Being a Data Scientist, getting updated with the latest research is very important since AI research is moving very fast. Reading research papers is not everyone’s piece of cake. It takes time to understand and to implement especially.
5. What advice would you like to give to DS starters or DS transitioners?
Do not skip mathematics and get aware that data science is applied mathematics only where you can’t escape from it.
MOOCs are not sufficient, take university classes from YouTube i.e. CMU 10–601,10–701 by Tom Mitchel/Alex Smola or CPSC-540/340 by Nando De Freitas.
Get used to LinkedIn, it’s a great tool for getting aware of your area of interest and may help you land our first job, that’s how I got mine.
Becoming a data scientist might seem lucrative, but it can be difficult and boring depending on your passion and interest.
So if it’s your passion then — learn the basics. Get a good understanding of statistical and mathematical concepts as they help in laying a solid foundation in the machine learning models you will be building.
Avoid running after certificates, get hands on. I have more than 25 certificates, no one asks for it in industry. Your work speaks for you.
Thank you for reading my post. I regularly write about Data & Technology on LinkedIn & Medium. If you would like to read my future posts then simply ‘Connect’ or ‘Follow’. Also feel free to listen to me on SoundCloud.