Varun Lakshmanan
Who I am
Hey! I'm Varun, a sophomore studying computer science at Georgia Tech. I'm a machine learning enthusiast interested in entrepreneurship and passionate about applying AI to solve real-world problems. In my free time, I compete in hackathons and work on personal projects, some of which have been internationally recognized.
Technologies that I work with include Python, Java, C++, Swift, MATLAB, TensorFlow, Keras, Scikit-learn, and OpenCV.
What I've done ✅
Experiences
Undergraduate Researcher, SALT Lab at Georgia Tech
August 2020 - December 2020
I worked as an undergraduate researcher under Dr. Diyi Yang.
- Worked on creating a model to automatically select the best data augmentations given a text dataset and task
- Tested and implemented different automatic data augmentation selection strategies for a classification task on multiple datasets
Software Development Engineer Intern, Amazon
May 2020 - August 2020
This summer, I'm interning at Amazon as 1 of only 100 college students to be recognized as an Amazon Future Engineer in the country.
- Working in the Paragon Frontend team, which creates tools for Amazon's customer service associates
- Creating an improved bookmarking function that allows associates to store bookmarks in a folder hierarchy, search and sort their bookmarks, and maintain 100% availability of bookmarks
- Using Java and AWS DynamoDB for the bookmark storage and retrieval and JavaScript, Vue.js, and BootstrapVue for the frontend to create a solution that reduces API calls by 50%
Research Intern, Scripps Research Institute
June 2018 - August 2018, June 2019 - August 2019
I interned two years at the Tomchik Neuroscience Lab at the Scripps Research Institute in Jupiter, FL. In my time there, I
- Deployed a computer vision system built in Python and MATLAB that could automatically classify different types of behaviors in fruit flies based on videos of fruit fly movement
- Standardized the use of a MATLAB toolbox to resolve errors in the computer vision system
- Developed an automated pipeline to create dynamic 3D visualizations of fly movement data by processing images in a program built in Java, preprocessing numerical data in VBA, and creating the 3D plot in MATLAB
Projects
BizViz
October 2020
BizViz is a financial business visualization tool that gives small business owners insight into different ways to keep their business operating during difficult times like the COVID-19 global pandemic. Instead of having businesses collapse due to owners continuing to spend more money to keep themselves growing, BizViz takes that same monetary value and instead shows what would be a reasonable return if that money was invested into different types of stock market portfolios (calculated using Facebook Prophet), as well as what the projected profit the business would have over that same time period (calculated by Sci-kit learn models, XGBoost, and LightGBM). All this data is then presented to the user in the form of a dynamic graph (built in React Native) to allow them to better gauge whether it would be worthwhile to divert funds into investments while the pandemic is active.
This project was recognized as being one of the top 8 projects out of more than 175 projects at HackGT 7.
Glass
May 2020 - June 2020
Glass is a Python framework that simplifies the process of building machine learning models for numerical data using Sci-kit learn. Given a dataset, the framework will automatically initialize a variety of classifiers/regressors with different architectures (Random Forest, SVM, XGBoost, etc), optimize all of their hyperparameters, and rank all of the models based on their accuracy against the dataset. The most accurate models will automatically be combined in an most optimized model ensemble possible.
COVID-19 Survival Calculator
March 2020
As part of a team in the COVID-19 Global Hackathon, I built a calculator that can estimate a COVID-19 patient's chance of survival based on about 10 inputs (gender, age, days before hospitalization, preexisting medical conditions, location, etc). I specifically helped preprocess our dataset, and trained an XGBoost model and LightGBM model on the data. I optimized the hyperparameters in both models and integrated them in a weighted model ensemble to maximize accuracy, which exceeded 95%. I used Python for this project.
Diagnosis of Cardiovascular, Renal, and Skin Diseases through a Computer Vision Model Ensemble in an iOS Application
August 2017 - March 2019
Over two years, I built and refined an iOS application that diagnosed serious heart, kidney, skin, and liver conditions based on a picture of your fingernails. I performed data augmentation to increase my dataset's size, then center-cropped and preprocessed the images using OpenCV functions. I built a convolutional neural network using Keras and ensembled it with several classifiers from the Turi Create framework. The model ensemble was integrated in an iOS application with a cohesive interface built using Swift.
I used Swift, Python, OpenCV, TensorFlow, Keras, and Turi Create for this project. For my work on this project, I was recognized by the International Science and Engineering Fair (Special Award from GoDaddy), ACM MobiCom (2nd in App Contest), Sigma Xi (1st in Computer Science Section), and the National Junior Science and Humanities Symposium (National Finalist and 1st in Florida).
What I'm working on 🛠
I'm planning to return to Amazon this summer. In the meantime, I'm doing independent research into reinforcement learning.
Contact me 📧
I'm always excited to explore new topics and work on new projects. Feel free to get in touch.