Projects

A three-class model predicting whether a student will pass, sit mid-band, or fall at risk in a module, replacing a vendor system, with leakage-safe validation and a hard focus on class imbalance.

A self-serve platform letting non-technical staff query institutional data in plain English through Databricks Genie; my work is the layer underneath: the curated tables and the model instructions that keep answers grounded.

Inherited a bloated, unstable Power BI model behind APAC's executive reporting and rebuilt it end to end: consolidated sources, modular SQL, row-level security, and pages built around each function's needs.

A market-by-market quarterly sales forecaster for 15 APAC markets, rebuilt from an abandoned proof of concept, where most of the work was redesigning the ingestion pipeline feeding the model.

An agentic pipeline that turns a partner organisation's 28-question survey into a formatted gap-analysis report, replacing a manual process that took one to two days per company.

An attrition study across seven years of student records that gave senior leadership its headline finding: Year 1 is where the large majority of departures happen, and the highest-impact place to intervene.

Scoped and drafted the statement of work for a vendor-delivered Fabric IQ pilot, choosing which dashboard cluster to include by reasoning from a lecturer's day-to-day query needs.

The region's general-purpose scanner-usage dashboard, built from nothing: with no dataset, I borrowed a scanner, traced where its activity landed across back-end systems, and assembled the source data myself.

A SQL classification model that sorts every Invisalign-submitting doctor in APAC by how they submit cases, so sales and marketing can target the ones likeliest to switch to iTero.

A cohort analysis of a scanner trial programme that isolated its real effect on Invisalign volume, comparing doctors six months before and after with a custom window keeping every comparison like-for-like.

A macro-driven Excel calculator for iTero scanner deals: enter a market and configuration, get revenue, cost, and margin net of every promotion and discount, without a trip to finance.

A Python and Selenium scraper that built regional lead lists for sales from public dental directories and competitor sites, proving out targeted lead generation instead of manual searching.