Inventory AI & Logistics

In the restaurant industry, forecasting is traditionally done manually at the end of the week. By the time a manager realizes they have a surplus of perishables, the window to reroute that inventory has often closed.

The client's operations team was spending roughly 40 hours a week attempting to manually coordinate surplus transfers between 50+ locations, leading to immense human error and inevitable food waste.

We moved the client from a batch-processing model to an event-driven architecture using Kafka. Every time an item was sold at any POS terminal, the central brain updated its localized demand curve.

We trained a lightweight XGBoost model that runs inference every 15 minutes, predicting end-of-day surplus with 92% accuracy based on historical sales, localized weather data, and active foot traffic patterns.

Prediction is useless without action. Once a surplus was flagged, our Node.js orchestrator automatically fired off API requests to secondary markets (like Too Good To Go) or arranged internal logistics transfers, completely removing the human manager from the loop.

The result was a 35% reduction in physical waste and a massive reclamation of managerial hours.