Portfolio
A selection of open-source projects, competition solutions, and research implementations.
Predicted severe COVID-19 outcomes from CT images (largest CT dataset of COVID-19 patients at the time). Used 3D DenseNet with augmentation on Tensorflow.
View on GitHub1st Prize winner of the Deakin Simpsons AI Challenge 2021. Deep learning classifier with extensive image augmentation and WandB experiment tracking.
View on GitHubVoice-based interface to control a smart home using speech recognition + encoder-decoder RNN to translate natural language into CDQL queries. 93% accuracy, 0.02% WER.
View on GitHubImplemented Deep Speech 2 for mission-critical ASR applications. Published in the official Keras example library. Analysed the effect of wartime noise on speech recognition.
View on GitHubIntegrated DALL-E into a voice-based UI so users can generate images from oral descriptions. Full-stack: Flask backend, React frontend.
View on GitHubFatigue crack detection in thin-rim helicopter planet gears for the 13th Defence Science & Technology International HUMS Conference.
View on GitHubETL pipeline and visualisations for pedestrian foot traffic across Melbourne to assist business location decision-making.
View on GitHubCredit scoring competition solution using LightGBM with feature engineering and WandB experiment tracking.
View on GitHubImplementations of deep learning fundamentals and cutting-edge techniques, based on the Coursera Deep Learning Specialization.
View on GitHub