Projects

This is a city-mapping software developed using C++, STL and data from OpenStreetMapAPI. The map is responsive; the level of detail shown is adjusted based on zoom and window size. Dijkstra's and A* search were used for routing between two points. Developed with two other team mates using Git for versioning.

This is a pathfinding visualizer of A* search. It was coded using Python's Pygame library, using its drawing methods to create the layout. The user denotes the start and end points with the first two clicks. Then, the user can draw blockages which are dynamically drawn in black, and pressing space will activate the A* search. Click on the image for a demo.

This is a CV project on gender recognition. A Mini-Xception model was implemented and trained using an IJB-C image dataset. Hyperparameters were tuned to improve performance, including batch size, lr, # conv filters and # epochs, resulting in around 6% improved accuracy. Downsampling to reduce class imbalance also resulted in a 9% performance boost.

Developed an XGBoost-based model to predict the onset of septic shock in patients 4 hours from time. Septic shock was defined according to an int'l standard and patient data consisted of chart data on their vital signs. Data was pre-processed from the MIMIC-III database.

Coded a neural network from scratch in order to understand how it works under the hood. Coded layers, activation functions, backprogagation, regularization, dropout etc. to complete the model.