In the Amazon Personalize console or via the SDK, create a dataset group of type DOMAIN with the E-COMMERCE domain; create an Interactions dataset and import your historical purchase/view events as CSV from S3 using CreateDatasetImportJob.
Create a recommender by calling CreateRecommender, specifying the datasetGroupArn, recipeArn (e.g., the 'Recommended-For-You' e-commerce recipe), and a recommenderConfig with minRecommendationRequestsPerSecond.
Wait for the recommender to reach ACTIVE status by polling DescribeRecommender; initial training can take tens of minutes to several hours depending on data volume.
Call GetRecommendations (personalizeruntime.GetRecommendations) with the recommenderArn and userId to retrieve a ranked list of recommended product IDs; optionally enable metadata in the response by setting enableMetadataWithRecommendations and specifying itemAttributes.
To enable real-time adaptation, create an event tracker (CreateEventTracker) and send behavioural events via PutEvents; Amazon Personalize updates user profiles for the next request within seconds for recipes that support real-time personalisation.
Known gotchas
Domain recommenders for e-commerce (using prebuilt recipes like 'Recommended-For-You') require a minimum volume of interaction events before they will train — too-sparse interaction data results in a training failure or low-quality recommendations.
Recommenders incur cost based on minRecommendationRequestsPerSecond even when idle; set this value to the minimum that meets your SLA and scale via UpdateRecommender if traffic grows.
GetRecommendations only returns item IDs; your application must look up product details from your own catalog database or feed — Personalize does not store or return product metadata by default unless you opt in via enableMetadataWithRecommendations.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp