I worked on the following two projects:

• Enhancing the accuracy of GNSS recievers in smartphones for better geospatial positioning of drivers
• Vehicle inspection system using instance segmentation and severity classification of damages, semantic segmentation of vehicle parts, and vehicle body style classification

The GNSS project needed deep domain knowledge which pushed me to learn numerous new concepts and strengthen myself in others including: Kalman filters, nonlinear optimization, how satnav systems work from a physical prespective using pseudoranges, signal identification by PRN codes, Keplerian elements transmitted by satellite messages, frames of reference, and many more.

For the vehicle inspection system, I worked on different steps of the ML pipeline from modeling by fine-tuning pretrained Mask-RCNN and ResNet-50 models in Tensorflow, to the optimization of the model training loop, to the preprocessing and merging of different datasets using custom scripts, and finally, to the manual annotation of hand-picked data for testing