What changed
AWS announced a multi-agent system designed to assist aerospace engineers in evaluating part substitutions. The system leverages Amazon Bedrock AgentCore, Amazon Nova Multimodal Embeddings, and the Strands Agents SDK to automate evidence assembly for substitution decisions. The solution includes a planner that routes tasks to three specialist agents, which search a multimodal knowledge base, score candidates against tolerance rules, and compare part photos. The engineer remains the final decision-maker, as the system drafts justification memos and Integration Impact Assessments without recommending or approving substitutions.
What this means in practice
The system is built around the principle that engineers make all decisions, with the system assembling evidence for review. It uses deterministic spec-fit evaluation through Python code to score candidate parts against reference parts and stated requirements. Multi-agent orchestration routes queries to specialist agents (Catalog, Evaluation, Knowledge) and synthesizes findings when multiple specialists contribute. The system also generates audit trail entries aligned with AS9100 standards. For image-based queries, the agent accesses an Amazon S3 Vectors index directly, bypassing the managed knowledge base retrieval API. The solution includes a fixed chain that produces formal Integration Impact Assessments (IIAs) and is deployed on Amazon Bedrock AgentCore runtime.
Limitations and availability
The solution requires an AWS account with access to Amazon Bedrock in the US East (N. Virginia) region. It also needs model access to Amazon Nova Multimodal Embeddings, Claude Haiku 5.5, Claude Son, and Claude Opus 5.5. A managed knowledge base created with Amazon Bedrock Managed Knowledge Bases using Nova Multimodal Embeddings is necessary for text and document retrieval. An Amazon S3 Vectors index over part photos and datasheet chart crops is required for image-as-query functionality. The system is available through the Amazon Bedrock AgentCore Gateway with the Web Search connector and a managed knowledge base target. Python 3.10 or later and Jupyter are required for running the notebooks. The complete code sample is a set of five Jupyter notebooks in a GitHub repository.
