Retail Demand Forecasting Platform

Replaced a spreadsheet-driven planning process with a probabilistic forecasting platform. Modelled demand at the store-SKU grain with a temporal fusion transformer, then reconciled forecasts up the product and region hierarchy so store, category, and national numbers always agree. Backtested against three years of sales, quantified prediction intervals for safety-stock decisions, and exposed the outputs through a self-serve dashboard the planning team owns.