Selected Work
Four projects that cover the range of what I do: computer vision, an end to end modelling pipeline, a deployed application, and a collaborative visualisation build. Each one has a full write up covering the problem, the approach, and what came out of it.
Do Discounts Affect Customer Ratings?
An end-to-end e-commerce analysis examining whether discounted purchases are associated with different customer ratings. I built the analytical dataset from relational Olist data using SQL and DuckDB, estimated the historical relationship with fixed-effects regression, and designed a randomized A/B test to answer the causal question prospectively.
View Project →
Bird Species Identification
Rather than reaching straight for a neural network, I ran four different families of method against the same bird image dataset: dimensionality reduction, classical statistical classifiers, density estimation, and convolutional neural networks. The interesting part is where the simpler approaches held their own and where they fell apart.
View Project →
Image Compression using ML
K-means clustering applied to pixel colour data to shrink an image's storage footprint. I took it past the notebook stage and shipped it as a live Streamlit app, so anyone can upload a picture, compress it, and download the result and see the tradeoff between file size and quality for themselves.
View Project →
SafeCity
An interactive tool built with a team to help prospective renters and buyers compare Los Angeles neighbourhoods on safety. Users weight 10 crime categories according to what actually matters to them, and the map ranks areas accordingly. Documented in a full paper and conference poster.
View Project →