At Sow and Soul, I design relational databases that turn farm animal records into a usable analytical asset for its Livestock Operating System. I also engaged with a South African livestock farmer to secure a letter of intent for pilot onboarding.
At Direct Line Group, I refactored and tested Semantic Role Labelling neural networks on size-variant driving data, achieving above-chance results (+50%). I integrated MLflow experiment tracking and model registry into the machine learning pipeline.
For my PhD research at the University of Hertfordshire, I developed CNN, LSTM RNN, and autoencoder models to classify aerosol particles from 2D fluorescence spectral data, achieving 82–97% accuracy for silica/coal discrimination. I also applied autoencoders to unlabeled UVLIF data for unsupervised feature learning.
Earlier, at Absa Group, I built an NLP tool to classify Twitter sentiment toward the brand and a help-desk query categorisation tool in R and Python. My work spans machine learning, statistical analysis, and data tools including TensorFlow/Keras, PyTorch, SQL, and AWS S3.
