What I led
Owned modeling and operations tooling end-to-end, from demand-forecast model development to shipping/uptime monitoring workflows used in day-to-day planning.
Inventory Forecasting Models
Supported inventory planning and route optimization for a large distributed retail footprint (700-800 locations) with mixed shipping modes (LTL freight pallets plus next-day ground), where on-time delivery and stock availability were both critical.
Stack
RShinyLinear forecasting modelsGradient boosted tree modelsLinuxData engineering/ETL collaboration with DBAs
Highlights
- Built and iterated demand-forecast models for a rotating-shelf problem where item-level history changed frequently and demand had to be inferred from store/location context.
- Progressed modeling strategy from linear approaches toward XGBoost as data volume/history matured.
- Implemented multi-horizon forecasting: daily short-horizon (5-14 days), placement-window horizon (1-2 weeks), and high-level location planning horizons.
- Built a Shiny-based management console to monitor location-level inventory, demand, shipping progress, estimated delivery, cost, and uptime in one workflow.
- Integrated forecast outputs into route and carrier decisions across mixed fulfillment modes (LTL and next-day ground).
- Coordinated closely with database and data-engineering teams to maintain inference-ready data pipelines.
- Needed to forecast demand for shelf-space contexts rather than stable SKUs, creating high volatility and weak early baselines.
Outcomes
- Supported large-scale fulfillment operations across approximately 700-800 locations with forecast-informed replenishment and routing decisions.
- Targeted approximately 95% uptime/no-stockout service levels for inventory availability.
- Improved shipping efficiency and service reliability across 700+ locations with forecast-driven routing.