Welcome!!

LinkedIn GitHub Twitter Kaggle Email

I am currently working as a Junior Research Fellow at IIT, Delhi. My current research work involves application of deep learning techniques on glaucoma detection. I am working in collaboration with AIIMS, Delhi. I also have past research experience at StellarDNN Lab where I worked on application of AI on astronomical data.

I also have experience working in industrial and teaching environments in the past. In my industrial role at Athenahealth, I primarily worked on test automation. My teaching role was at Univ.AI where I assisted in developing AI courses.

My fascination in AI grew as I wanted to decode the human mind. This is also one of the motivating factors to keep the grind going!!

In my free time, I watch football and I do like watching TV shows.

Neurocomputing Lab, IIT Delhi, India
Junior Research Fellow

Auguest 2023 - Present

  • Leading the development of a deep learning model for predicting visual field charts from eye-tracking sequential data retrieved from normal and patients with glaucoma.
  • Conducted analysis of eye tracking data, employed interpolation methods for handling missing values, and used filters for improving data quality. Developed an initial prototype model combining both language and vision models.
  • Currently experimenting with transformer architecture on this use-case.
  • This is being done using Pytorch.

StellarDNN Lab, Harvard
Research Assistant

Sep 2022 - Dec 2023

  • Conducted an in-depth, code-centric analysis of the research paper: Astromer: A transformer-based embedding for the representation of light curves.

  • Conducted analysis of spectral data consisting of a sequence of pairs of flux and magnitude derived from stars, galaxies and quasars. Employed data visualization techniques using Matplotlib leading to insights that prompted improvements in the data-gathering process. Performed correlation and autocorrelation study to check the applicability of an autoregressive process on spectral data. This justified the application of Astromer on spectral data.

  • Developed a data-informed masking technique as part of the data-preprocessing pipeline of Astromer. This technique focused on the emission and absorption region of a spectrum which is in departure from the technique used in Astromer. ◦ Enhanced the transformer architecture of Astromer resulting in improved performance in processing spectral data. ◦ Leading the development efforts on this research project aimed at deriving spectral data embeddings from stars, galaxies, and quasars.

  • This is being done using TensorFlow.

Univ.AI, Bengaluru, India
Teaching Assitant

May 2022 - Dec 2022

  • As a teaching assistant, I have helped in course content development for the following courses: AI Basics , Data Science Basics , Convolutional Neural Networks , Language models.

  • My role involved the following:

    1. Content development (theory)- Addition and simplification of concepts for more efficient delivery.
    2. Content development (practice)- Refinement of labs, exercises related to theory sessions.
    3. Doubt clearance- Answering student doubts on forum (ed), guiding students during lab hours.

Athenahealth, Bengaluru, India
Associate Member of Technical Staff

August 2021 - July 2022

  • I worked as part of test automation team in which I contributed in writing test automation scripts for:

    1. Forms meant to be used by providers to collect health information from patients.
    2. Workflows to keep track of the vaccination history of patients.
    3. Workflows to add medications to a patient’s electronic health record (EHR).
    4. Workflows to help providers generate medical bills for patients.
    5. Workflow to refer patients to other providers for further treatment.

  • This work was done using C#.

Athenahealth, Bengaluru, India
Intern

January 2021 - July 2021

  • I contributed to the migration of performance engineering scripts from LoadRunner to JMeter for both: athenaPractice and athenaFlow.

RVCE, Bengaluru, India
Bachelor of Engineering in Computer Science

2017-2021

  • Aim: This project aims to implement Vision transformer paper from scratch in Pytorch.

  • Motivation: The motivation behind this project was to get a deeper understanding of vision transformer architecture. It was also intended to impart a better grasp on replicating research papers.

  • Dataset: MNIST.

  • Improvements: Additional things to add: W&B integration, aim for better test accuracy, and code optimization.

  • Note about implementation: This implementation was solely meant for learning purposes. This is reflected in the comments supporting lines of code.

Astrospec

Ongoing

  • Aim: To extract general representations of millions of stars, galaxies and quasars by training a self supervised transformer: ASTROMER.

  • Dataset: Custom dataset containing stars, galaxies and quasars from SDSS DR17.

  • Significance: The representations obtained from ASTROMER model speed up training of downstream tasks such as classification of variable stars, galaxies and quasars.

  • About: This project was aimed at exploring data, doing a comparative analysis of ML models based on their ability to correctly classify churn customers.

  • Importance: Businesses want to maximize their revenue. It is generally easier for a business to retain customers than to onboard new ones. To grow their revenue, businesses must target new customers but also make sure that churn is minimized. If they can identify potential churn customers, they can run programs to retain them.

  • Dataset: In this project, an imbalanced dataset is used. This closely mimics the real world.

  • Algorithms used:

    • Categorical Naive Bayes
    • KNN
    • Logistic Regression
    • Random Forest
    • Decision Tree
    • Decision Tree with Bagging
    • Adaboost
    • XGBoost.

  • Metric used: F1-score is used because we are dealing with an imbalanced class classification problem.

  • About: This project contains some of the most popular algorithms implemented in python only. I took inspiration from open source code repositories to code these algorithms.

  • Importance: It is always important to code things from scratch to develop a deep understanding of concepts. I plan to continue adding algorithms to this project.

  • Algorithms implemented:

    • KNN
    • Logistic Regression
    • Linear Regression
    • Decision Tree
    • Kmeans clustering