Technical Approach
Architecture
Data ingestion from ERP → Feature store → ML forecasting service → Replenishment engine → ERP
Models
Baseline: SARIMA
Production: Gradient Boosting / LSTM (SKU-level)
APIs
REST (JSON)
Batch + near-real-time sync
Complexity
Forecasting: O(n × t) where n = SKUs, t = time windows
Trade-offs
Accuracy vs. interpretability
Model complexity vs. operational stability