Researching Imitation learning techniques for Task and Motion Planning Problems (TAMP) using Hetero-GNNs

Developed the first ever (suite of) model(s) for gap junction annotation for connectomes. Developed model decreased annotation time by over 150x.New pre-processing & post-processing methods for raw Electron-Microscopy images of the C. elegan’s worm, Hyperparameter search on focal loss, custom loss functions based on focal loss and dice loss, custom inference pipeline, and customizations to the Unet architecture led to further recall of 80% at 70% precision .

Responsible for a team of 6 technical writersKeeping up to date with the academic research space in machine learning, reading papers, and synthesizing them for the monthly UTMIST newsletter -deMISTify

Responsible for conducting student office hours, review sessions, tutorials, and more; enhancing student comprehension during and outside of instructional hours.

Inferring spatiotemporal urban vegetation data in Toronto from traffic cameras, phenocams, and aerial surveys using Computer Vision for better carbon sequestration estimates in Toronto

Working with Professor Debra Wunch (Atmospheric Physics) and her team in characterizing, monitoring, and quantifying Toronto’s urban emissions. I deal with data from satellites and various ground-based sensors like EM27-SUNs and those of the Pandonia Global Network.

Develops apps to enrich student life and engagement on campus for the University of Toronto