Detection of Thoracic diseases through Chest Xrays

• Dataset: NIH Xray images. 112,120 scans of chest (frontal view). 14 different classes (diseases). Bounding box ground truth for a subset of images.
• Models: MobileNet, VGG-16, DenseNet121 (or CheXNet) to identify the classes, and GradCAM for activation maps.
• Tools and Technologies: Tensorflow with Keras, matplotlib, seaborn, pandas and numpy for Data Analysis. Deep Learning VM on Google Cloud.

COVID-19 Global Forecast

• Dataset: Novel Coronavirus cases (JHU)
• Models: Exponential, Logistic, Polynomial, Ridge and Lasso regression, SIR model
• Tools and Technologies: R, Rstudio, Python

Machine Reading Comprehension

• Dataset: SQuAD
• Models: BiDAF, BERT and ALBERT
• Tools and Technologies: PyTorch, Google Colab

Taxi Fare Prediction

• Dataset: New York City taxi fare prediction dataset
• Models: Linear Regression, Polynomial Regression, Random Forest regression
• Tools and Technologies: pandas, numpy and scikit-learn

CVTetris

A cooler version of the popular Tetris game which allows the user to control the falling tiles using hand motion. I worked on the main logic, multiple object detection, optimization, reliability and UX of the game. I used the openCV and Pygame libraries. Also this was a top 15 hackathon winner.

Geotroner

This was my first ever (successful) hackathon project. I worked on the front-end development. I learned a lot of new concepts and tools through this project, like JavaScript, CSS, Bootstrap, UI designing and user authentication. Also this was a hackathon winner amongst 200 teams .

PlanItRight

As part of my independent summer development project, I made a personalized day-to-day itinerary planner from scratch. Through this project I learnt a lot of fundamentals of web development and software engineering, as well as NodeJS, Express and web scraping.

Note: Some features of the app in the demo have been disabled due some issues.

Infinity File

The idea behind this project was to use Imgur's image upload API without any limits on number of images uploaded to create a theoretically safe and infinite file storage system by converting any file type to an encrypted image.