This article describes how Amazon uses Amazon Bedrock to scale Rufus, its AI-powered shopping assistant, to millions of customers.

• Rufus serves 250+ million customers with 140% YoY monthly user growth and 210% YoY interaction growth
• Customers using Rufus are 60% more likely to complete purchases
• Amazon transitioned from custom in-house LLM to multi-model approach using Amazon Bedrock
• Development velocity increased 6x by using Amazon Bedrock's managed infrastructure
• Rufus uses multiple models including Amazon Nova and Claude Sonnet for different query types
• Amazon Nova Web Grounding reduces answer defects through web search and source citation
• Prompt caching and parallel tool calling optimize latency
• Agentic features include price history, price alerts, auto-buy, and conversational reordering
• Auto-buy customers save average 20% per purchase
• Amazon Bedrock handles global deployment across multiple regions

Amazon demonstrates how Amazon Bedrock enables rapid AI application scaling through flexible model selection, managed infrastructure, and agentic capabilities for production deployment.