Analytics, machine learning, and side projects from six years of commercial practice and personal exploration.
An end-to-end ML product for identifying at-risk telecom customers before they leave, from raw data exploration through to a deployed Streamlit prediction app. Built on 594K synthetic training rows from the IBM Telco dataset. The key finding: cost-sensitive modelling delivers 25x more business value than a standard classifier at identical AUC.
Three years of transactional data from a fictional Australian online grocer, dissected across 16 business questions using advanced SQL. Covers cohort retention, RFM segmentation, promotional ROI, market basket analysis, and category performance. The full analytical toolkit of a commercial analyst.
A free, privacy-first Chrome extension for blocking distracting websites. No account, no subscription, no data collection. Just configurable blocking with Pomodoro timer, day/time schedules, named block lists, and hard-lock protection. Built because every decent option on the Chrome Web Store either costs money or phones home.
An interactive classroom tool that lets students navigate a real loss surface in 1D and 3D, watching SGD, Momentum, RMSProp, and Adam converge (or fail to) as they tune the learning rate and schedule in real time. Built as a teaching aid for UTS data science students.
A Power BI case study for a fictional Australian e-commerce brand, built on a star schema data model. Demonstrates DAX measure design, data modelling decisions, and business storytelling using a structured hypothesis-driven narrative.
A predictive scoring system built to proactively identify at-risk key accounts during COVID-19, enabling account managers to intervene before contracts were lost and achieving 100% Key Account retention across the portfolio.
A geospatial analytics project that audited national field territory allocation, identified distribution inefficiencies invisible to conventional reporting, and redesigned routes to deliver a 16% reduction in Route-to-Market costs.
Design and automation of a multi-report Power BI suite that replaced fragmented manual reporting across 300,000 retail outlets, resolving longstanding data discrepancies between field systems and head office.
An unsupervised ML pipeline combining K-Means clustering with statistical outlier detection to identify non-compliant sales behaviour across a 300,000-outlet retail network, protecting incentive payouts and strengthening reporting integrity.
A CNN-based model that predicts the age class of a person from a photo. Facial regions are isolated with MTCNN and cropped with OpenCV before age prediction is applied — inspired by working at Dan Murphy's and checking customer IDs.
A predictive model built to tackle booking cancellations in the hotel industry, driven by the shift in customer behaviour through online reservation channels. Evaluated multiple classifiers with a focus on interpretability and commercial applicability.
A Tableau dashboard analysing Australia's Machinery and Transport Equipment trade from 1988 to 2022, using data sourced from the Australian Bureau of Statistics (ABS). Covers import and export trends, trade balance shifts, and economic context over three decades.
No projects in this category yet. More coming soon.
