Outcome first. Each of these started with a business question and ended with a number someone could act on.
+$26,477 / yr
Swire Coca-Cola: growth optimization
Found an optimal 300-gallon threshold through ETL pipelines, sensitivity analysis and predictive modelling. Raised the growth-ready customer share from 4.12% to 5.27% and built Tableau dashboards to balance growth against delivery cost.
PythonSQLTableauETL
98% AUC · 0.91 recall
Home Credit: loan default risk
Predicted loan defaults with a tuned gradient boosting model on 37.5M+ records. Feature engineering, upsampling and hyperparameter tuning to cut losses from defaulters classed as creditworthy.
PythonGBMFeature engineering
$26,559 less pricing error
Ames property price prediction
Neural-network price model for Ames houses. Outlier detection and clustering (PCA, k-means) to find the pricing segments that matter to buyers and investors.
PythonNeural networksPCAk-means
175K tweets analyzed
Super Bowl game day challenge
Sentiment analysis and time-series forecasting on tweets about 2024 Super Bowl ads. Identified the top commercials, ad-timing effects and viewer preferences, reported as advertising ROI in Tableau.
RSQLPythonTableau