Aditya Shaha is a Data Science Engineer at Blinkit, working in the personalization team. Before joining the personalization team, he worked on multiple problems on the supply side of Blinkit. In his free time, he loves reading books and listening to podcasts. Here is Aditya sharing his experiences and learnings so far at the heart of instance commerce.
Tell us something about yourself and your journey in Blinkit so far.
I am not sure if I have something pithy which introduces me well, but If I had to, I would love to describe myself as a “curious learner.” I guess those early years of watching my grandfather dismantle numerous machines and trying to fix them just for fun inadvertently imbibed this trait in me. As a data scientist, I find this trait quite helpful as deriving insights from messy data demands your curiosity and a learner’s mindset.
I began my journey at Blinkit almost 3.5 years ago, as an intern in the Data Science and Engineering team, even before graduating from college. In college, my exposure to the field of Data Science had primarily been through the lens of Kaggle competitions and research work. Most of these problems were already distilled into well-defined data science problems. I soon came to realize that being a data scientist in a startup, scaling up its Data Science team, this is not always going to be true. However, this lack of definition, forced me to work on data problems from their ideation to production, which painted a much more holistic picture of how data science can help bring value in the real world.
If you were to describe your experience in Blinkit in one word, what would that be?
On the technology side, what has been the single greatest change you’ve witnessed in the organization over the years?
I am not sure if there was a single change that created a huge impact, but the process of the organization moving from simple average-based models in excel sheets to building sound data models based on org-wide accepted sources of truth was quite pivotal.
I was fortunate enough to join the Blinkit data team at a relatively early stage. It gave me an opportunity to witness the entire lifecycle of a data science project from a simple baseline model driven mostly by human intuition to a nuanced machine learning model with minimal human intervention. It also helped me realize the importance interpretability plays when the model predictions have 2nd and 3rd-order effects.
One of the initial projects that I worked on helped cement this idea better. I believe the success of that project can not only be attributed to the decent accuracy of the model, but the confidence that the stakeholders had in the model because of its interpretability. The interpretability aspect not only helped gain stakeholder trust on model predictions but it also helped us define better ways to remove spurious correlations and introduce better constraints on model outputs. So the system transitioned from a simple average-based baseline to an interpretable ML model with a human-in-the-loop system to finally going on an autopilot mode with minimal intervention. This process of introducing ML models by leveraging human intelligence to aid better decision-making, I suppose, has been quite seminal.
What are the bigger challenges data helps in solving in the e-commerce space?
One of the exciting parts of working in the Q-commerce space is that the space is rife with problems that can benefit from almost all paradigms of data science solutioning. May it be supervised, unsupervised, reinforcement learning or optimizing the existing processes, we have a use case for all. Along with the paradigms there is also a plethora of data types that one gets to dabble with ranging from structured tabular data, to unstructured image, text and voice data and also the GIS data.
This data can then be leveraged to build solutions which interact with the customers directly like Search, Recommendations, ETA estimation and complaints resolution or indirectly in optimizing the supply chain by accurate demand forecasting, store location scouting, store level assortment selection to name a few which help us bring the magic of delivery within minutes to all our customers.
As a techie, what is the principle you always adhere to while solving a problem statement?
Reasoning by first principles and not by analogy.
Who do you seek inspiration from on an everyday basis?
I enjoy reading books and listening to podcasts that fuel me with a timely dose of inspiration. But apart from that, the fact that I get to brainstorm on problem statements with a bunch of smart people, the solutions to which have a direct impact on millions of customers on a daily basis is what gets me up from my bed.
What do you love the most about your colleagues?
The people that I have had the privilege to work with are intelligent, exorbitantly passionate about their work and yet humble enough to consider everyone’s perspectives. Everyone is very approachable and open to collaboration which personally made it very easy for me to work on various problem statements when I was just starting out.
How do you strike a work-life balance?
I don’t do a good job at it. I have a tendency to get engrossed in the problems that I am working on, which make it a tad bit difficult to leave work at the door. However, I am trying to incorporate some physically intensive activity in my schedule like an evening run to help me bring out that balance.
What would be your advice to someone hoping to join Blinkit in the data space?
Data Science is not only about model fitting, at least at Blinkit. As a data scientist, you would be given the responsibility to drive a problem from its inception to its deployment. Taking up these responsibilities will not only help you become a well-rounded data scientist but also help you drive a lot of impact in the organization.
Communication skills are underrated. There exists a lot of knowledge in the system already. People on the ground can help you discover variables and interactions which siloed experimentation would take weeks. Investing time in your communication skills and learning how to collaborate well can turn out to be hugely beneficial in the long run.
Aditya Shaha is a Data Science Engineer at Blinkit. You can follow him on LinkedIn.
Thanks for reading this blog. Say hello on Twitter or follow us on LinkedIn.