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