Design an item category prediction system

Design an end-to-end ML system that predicts an **item’s category/type** (multi-class or hierarchical classification), e.g., assigning an e-commerce listing to a taxonomy like `Electronics > Headphones`. ### What to cover - Inputs available at prediction time (title, description, images, attributes, seller signals, etc.) - Training data and labeling strategy (including noisy labels) - Model choices (text-only vs multimodal; flat vs hierarchical) - Offline evaluation metrics and online success metrics - Serving architecture and latency/throughput targets - Handling cold-start items, new categories, and class imbalance - Monitoring, drift detection, and retraining/feedback loops - Human-in-the-loop / moderation workflow for low-confidence cases