- Analysed research papers and utilised machine learning techniques with PyTorch and Python toapply radar signals processing, improving accuracy from previous method by at least 10% - Reduced computing time and GPU memory usage by 3 fold via modelling of complex-valued andmulti-channel matrixes into smaller real 2-channel matrixes for neural networks input- Engineered CVAEs and ResNet models to result in true positive rate of 91%

- Engineered MRT crowd simulations in Python with NumPy and pandas to replicate ticketing data- Developed mathematical models using graph theory in NetworkX for application of algorithms- Employed numerous statistical and optimisation techniques, including kernel density estimation andmaximum independent sets, to infer overcrowding from MRT ticketing data- Designed quantifiable metric systems to evaluate performance of methods on real data, achieving aprecision of 86.6% and F1 score of 93.1%