Making an impact with code, I believe that is something that could summarize my undergrad years.
I’ve always loved Python because of the pace with which I could translate ideas into code. What started as a fixation with web scrapers soon morphed into a love for OpenCV. I was building Sudoku Solvers and Emotion classifiers soon, and I honestly had so much fun doing it.
It was around then that I came across a Chatbot Hackathon and I started tinkering with Natural Language Processing (NLP). I won the hackathon and in a month, I was building a chatbot from scratch to fulfill service queries within the GoBumpr’s app. It was stimulating to build programs that could think for themselves. And as automating stuff with Python became mainstream, I was now obsessed with building Machine Learning (ML) models to solve interesting problems. Later that year, I joined the team at BicycleAI where I helped them develop their models to predict the similarity between two questions using a BiLSTM-based model. This was my window to ML research since this was the first time I had to read papers and create my own model. With great models comes the need for great datasets; In order to train our model, I also had to develop a robust FAQ scraper. Life indeed is a circle.
I was smitten. The growing potential and groundbreaking research in Artificial Intelligence (AI) motivated me to work on a more challenging problem and I got an opportunity to do so at MSR’s 2018 Summer Workshop. Along with a team from IIIT Delhi, I worked on Compression of Neural Networks (CNNs) using Bayesian Inferences. Our project was divided into two parts– the implementation of CNNs on FPGAs and the compression of CNNs before deploying on the FPGA. I was responsible for the latter. Our work was derived from “Bayesian Compression for Deep Learning”, (NIPS 2017).
I have always been a believer in learning in a manner that is practical foremost, and this is evident from the immense amount of code I’ve written. I enjoy building something new with my mind. While I do someday plan to document my knowledge and experience in research papers, right now, writing great code gives me the most satisfaction.