Question

A media company is building a document-processing pipeline that must run three different Bedrock foundation models in parallel against the same input text (one for summarization, one for sentiment scoring, one for entity extraction), then merge the three results into a single JSON response. Each model invocation can take 10-40 seconds, and the pipeline must handle transient Bedrock throttling with automatic retries and backoff. The team wants a fully managed orchestration layer with minimal custom code for coordination and error handling. Which approach best meets these requirements?

Accepted Answer

Build a Step Functions Standard workflow using a Parallel state with three branches, each invoking Bedrock via the optimized service integration, and configure Retry policies with exponential backoff on ThrottlingException for each branch.

Answer

Write a single Lambda function that calls all three models sequentially using the Bedrock SDK and implements custom try/except retry loops, invoked synchronously through API Gateway.

Answer

Use a Step Functions Express workflow with a Map state that iterates over the three model IDs, relying on the default at-least-once execution semantics to handle throttling.

Answer

Configure a Bedrock Agent with three action groups, one per model, and let the agent's orchestration prompt decide the invocation order and merge the outputs.