Department of Computer Science, University of Wisconsin - Madison, USA
Contact Me Download CVMay 2022 - August 2022
Worked on VM-to-VM connection reliability in the Andromeda Software Defined Networking (SDN) Stack of Google. Added more than two thousand lines of code in the production code-base consisting of more than one million lines. Added 12 new features related to telemetry and test coverage. Introduced new monitoring capabilities and helped in early bug detection and resolution. C++ for core implementations and Python for visualization and monitoring.
March 2021 - June 2021
We developed an android app for on-ground surveys by agents, containing features like an attendance management system, online form submission with location tracking, reimbursement management system. The app is pretty smart to detect fraud cases, like using simulators with fake GPS to mark attendance. We took the help of the Safety Net Attestation API for this. The language and architecture used were Kotlin and Model-View-View-Model(MVVM) architecture along with View Binding for compile-time binding class generation of XML layouts. Network calls were made using Retrofit via HTTPS requests, and the Android Room Database for offline storage of data. Firebase Cloud Storage was used to store files like images in a systematic order.
June 2020 - July 2020
We developed a visualization platform of time-series data from Atlassian JIRA using Grafana. My job was to built custom visualizations as plugins for Grafana using ReactJS. The data was scraped using python scripts and stored in InfluxDB. This internship introduced me to the Agile Methodology of Software Development where we deployed regularly on the Continuous Integration Continuous Development (CICD) pipeline. The pipeline consisted of Atlassian Bit-bucket, Octopus, and Open-Shift. This experience also taught me Docker containers and Kubernetes, used during deployment in the CICD pipeline.
March 2019 - December 2019
I led a team of 10 students to develop a website for operation management of the Inter-IIT Sports Meet 2019 that happened in IIT Kharagpur. We designed the front-end using ReactJS and back-end using Django with MongoDB as database. The team also featured an android app which was made with Flutter, having similar features to that of the website. They featured registration of students and staff members, live-score of ongoing matches, automated ranking system, mess management of participants, etc. This was one of the first live web projects I had worked on, used by some thousands of students for conducting a successful Inter-IIT Sports Meet.
May 2018 - Till Date
We started an organization to help students of IIT KGP with previous year's question papers. Students find it difficult to adapt to the new academic structure as it is very different from the entrance examination. The JEE entrance examination is objective whereas IIT exams are mostly subjective. Thus, we decided to create a book containing the previous year's question papers and detailed subjective solutions. We started as a business venture where we extracted a revenue of around 2 lakhs in the initial year. Looking at its thriving success, we also started voluntary service, providing important study materials for free to students collected from toppers of their previous batches only for the freshman year. To know more, have a skim through this article.
August 2021 - Present
Advisor : Professor Paul Barford, UW Madison
We are doing empirical analysis of DNS structure, behavior and performance using novel data sets and techniques. The goal of this work is to develop broad and novel perspectives on DNS that lead to improvements in the systems and protocols used to configure, manage and secure the underlying infrastructure. This project is guided by Professor Paul Barford, UW Madison along with Professor Mark Crovella, Boston University and, Joel Sommers, Colgate University.
November 2020 - April 2021
Guide : Prof. Animesh Mukherjee & Prof. Saptarshi Ghosh from IIT Kharagpur
We studied the existence of bias of the Amazon buy-box winner algorithm towards Special Merchants like Cloudtail and Appario, and the promotion of Amazon private products on their website. We also submitted a publication in the ACM Conference on Computer-Supported Cooperative Work And Social Computing (CSCW) for review. This article from Reuters is very helpful to qualitatively understand the bias. Our analysis focussed on providing numerical to the idea described in this article with data scrapped from the Amazon website.
May 2020 - October 2020
Guide : Prof. Animesh Mukherjee & Prof. Mainack Mondal from IIT Kharagpur
Guide : Prof. Abir De from IIT Bombay
We developed a mobility-aware multi-agent simulation driven method for the estimation of COVID-19 spread in India (although the model can be applied for any country). Our method intrinsically took care of factors like mutation of virus strains, asymptomatic cases, and other latent conditions using a human mobility simulation consisting of 1.38 billion people in India. The aim was to estimate the amount of testing needed under different levels of infection. To know more about our analysis and research, please read this article.
May 2020 - October 2020
Guide : Prof. Niloy Ganguly from IIT Kharagpur
We worked on determining fairness in recommender systems using election voting methods and traditional recommender system algorithms like Matrix factorization on the Smart Media Adressa News Dataset. We applied the theory of electoral systems like Single Transferable/ Non-Transferable Vote, k-Borda count, Bloc Voting, etc. to measure fairness. Finally, we used Pointwise Mutual Information to extract the political bias of an article and devised a bias metric to measure the aggregate ideological bias of a recommended set of articles. The different electoral systems were compared for user satisfaction and bias as a result.
Publication : Two-Sided Fairness in Non-Personalised Recommendations
Conference : AAAI Student Abstract 2021
June 2020 - October 2020
Guide : Prof. Kaushal Kumar Bhagat from IIT Kharagpur
Guide : Prof. Nasser Giacaman from the University of Auckland
We designed a language model based on BERT using python-PyTorch and Transformers library to systematically attach sentiment labels (Positive, Negative, Neutral) and seven pre-defined aspect labels (Technical Issues, Usability, Content, User Interaction, Feature Request, Learning Qualities, Advert) to the scrapped app reviews from the Google Play Store. We then compared the user dispositions of apps from five different app categories (AR, VR, Educational, Educational AR, EducationalVR) to analyze the effectiveness and quality of AR/VR for education in the form of android apps compared to the traditional norms and techniques.
Publication : Analysing user reviews of interactive educational apps: a sentiment analysis approach
Journal : Interactive Learning Environments
January 2019 - January 2020
Guide : Prof. Dipankar Dasgupta from Memphis University
I explored evolutionary algorithms like Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization, etc. and scrutinized their plausibility to design neural network architectures. The first step involved devising gene structure to represent salient features of a Convolution Neural Network (CNNs) with Skip Connections. The second step included fine-tuning genetic operators like Crossover and Mutation to apply on the devised gene structure. The codes were on python-Tensorflow and the dataset used was MNIST handwritten digit data set, and Fashion-MNIST data set.
Publication : Evolution of Convolution Neural Network Architectures using Genetic Algorithm
Conference : WCCI 2020 and IEEE CEC 2020