Build an AI-powered product tagging system with Amazon SageMaker serverless model customization

This article demonstrates how to build an AI-powered product tagging system using Amazon SageMaker serverless model customization to fine-tune Qwen3-8B for consistent catalog enrichment.

• Customize Qwen3-8B with supervised fine-tuning (SFT) to teach the model a stable tagging schema
• Optimize with reinforcement learning using Group Relative Policy Optimization (GRPO) and verifiable rewards
• Use Amazon SageMaker serverless training to manage capacity automatically without selecting instances
• Deploy the optimized model to Amazon SageMaker Asynchronous Inference for batch catalog processing
• Design deterministic rewards combining recall, precision, accuracy, and formatting metrics
• SFT achieves 0.6827 overall score; GRPO further improves to 0.6941 with higher recall

Serverless model customization provides a managed path to specialize open-weight models for high-volume, repetitive tagging tasks with stable schemas and programmatic scoring.