Built a Retail Intelligence Platform using 540K+ UK e-commerce transactions, combining churn prediction, demand forecasting, and LLM-generated business narration.
My XGBoost churn model used SHAP explainability and cost-based threshold optimization, cutting projected business cost by 38% and identifying $394,552 in probability-weighted revenue at risk across 784 customers. I also developed Prophet forecasts for 20 products that outperformed a seasonal-naive baseline in 70% of backtesting folds.
I delivered the platform through a FastAPI backend with 15+ REST endpoints and a three-tab Streamlit dashboard. I also built a fully local Ollama/Llama3 and LangChain layer to explain predictions and draft retention emails without sending data outside the system.
