AWS builds engineering assistant for part substitution with Nova and AgentCore
AWS has detailed how its Amazon Nova Multimodal Embeddings and Amazon Bedrock AgentCore can automate evidence assembly for aerospace engineers evaluating part substitutions. The system uses three specialist agents—Catalog, Evaluation, and Knowledge—to retrieve specs, score candidates against tolerance rules, and compare part photos via visual similarity. A Claude Haiku 5.5 planner routes…
Key points
- Three agents—Catalog, Evaluation, Knowledge—use **Amazon Nova** for multimodal retrieval and **Bedrock AgentCore** for orchestration, with **Claude Haiku 5.5**, **Sonnet 5.5**, and **Opus 5.5** handling routing, execution, and synthesis
- Visual similarity scoring compares part photos via **Amazon S3 Vectors**, while **AgentCore Gateway** fetches live web data for lifecycle claims, with explicit fallback logging
- Deterministic Python code scores substitutions against tolerance rules, and the system drafts memos/audit logs but never approves substitutions, adhering to **AS9100** requirements
The solution embeds datasheets, charts, and photos into a single semantic space using Amazon Nova Multimodal Embeddings, enabling queries like ‘torque curve for the 0.06 Nms wheel’ to retrieve matching visual content. For image-based searches, it uses Amazon S3 Vectors alongside AgentCore Gateway for live web data checks. AWS highlights two retrieval paths: one for text/document queries via Bedrock Managed Knowledge Bases and another for image uploads. The system also logs silent fallbacks to avoid undetected errors. Deployment requires AWS GovCloud (US) for aerospace compliance and supports Strands Agents SDK for customization. The architecture is modular, allowing deterministic scoring tools to integrate with PLM or ERP systems without agents.
Model pages: Claude Haiku 5.5 → · Claude Sonnet 5.5 → · Claude Opus 5.5 →
The story so far
4 episodes →- AWS builds engineering assistant for part substitution with Nova and AgentCorethis story
Agentic engineering intelligence with Amazon Nova Multimodal Embeddings and Amazon Bedrock AgentCore
AWS Machine Learning Blog · 2 November 2026
Loading the full article…
This text was published by AWS Machine Learning Blog and written by Aishwarya Gavili. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Enterprise & Industry
All →- 42AI launches behavioral world model for predicting trader actions · 1 src
- Webb: Shadow AI as Workflow Issue · 1 src
- Snowflake report: 48% of agentic analytics insights improve business · 1 src
- Athenahealth links patient records to Anthropic’s Claude for Pro and Max users · 1 src
- Microsoft builds Windows 11 speed test in a day using agentic AI workflow · 2 src
Comments
via GitHub Discussions