Design
1 min read
Ai-muse
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
SF
Jan 05, 2026