The AWS Machine Learning Blog describes how to deploy a multimodal ordering assistant on WhatsApp using Amazon Bedrock AgentCore with Amazon Nova 2. The assistant accepts customer orders through text, voice notes, and real-time voice calls, all on a single business phone number. In the described architecture, the channel layer and the ordering layer remain separate, and a single shared memory recognizes each customer across all three input channels.
Why it matters
Consolidating multiple input types onto one number, with shared memory that tracks a customer across channels, addresses a common challenge in conversational ordering systems where interactions are otherwise fragmented across separate touchpoints.
Who should care
Developers and businesses building customer-facing ordering or messaging assistants on AWS, particularly those using WhatsApp as a channel, may find the described pattern relevant.