SWEET - Weakly Supervised Person Name Extraction for Fighting Human Trafficking

A weak supervision pipeline for extracting person names from noisy human trafficking proxy data (escort ads). Paper also presents HTGen, a new synthetically generated dataset of escort advertisements (built using GPT) to facilitate further research within the community. Code, Poster, Paper

Image Classification from Scratch And Scaling Using Multicore Matrix Multiplication

This is the project I took apart in when I was volunteering at the Prometheus Lab at McGill. The java vision code takes images from a robot and runs a custom image classification java program written from scratch to classify what type of room the robot is in. I implemented batching, serialization, and scaled the code to handle larger amount of training and testing data using multicore matrix multiplication packages. Poster

Brain Tumor Segmentation with Attention-Based U-Net

This project was the final project I did for a summer machine learning course I took with MIT Professor. Mark Vogelsberger which we later published as a full length paper. In this project we improved the existing U-Net model by adding additional Squeeze-and-Excitation Block and CBAM attention modules into the decoder blocks to improve the performance on the brain tumor segmentation task. Paper

Rocket Engine DAQ

For my internship at USC Liquid Propulsion Laboratory, I worked in the DAQ team and implemented python code that fetched and visualized engine data read from labjacks placed in the DAQ monitoring system.