AWS-powered Virtual Engineering Workbench cuts automotive test-case creation time by up to 80% using Amazon Bedrock
An AWS Professional Services team integrated Amazon Bedrock into the Virtual Engineering Workbench (VEW) for automotive software testing, reducing test-case creation time by up to 80% while maintaining accuracy through a human-in-the-loop approach. The solution was implemented in production in four weeks.
Background
Modern automotive software programs require managing hundreds of thousands of requirements, each needing verified test cases. The manual process is slow, repetitive, and difficult to scale as vehicles become more software-defined. AWS Professional Services identified test-case generation as a high-value automation target within the Virtual Engineering Workbench.
What Was Implemented
- AI-powered extension of the Virtual Engineering Workbench (VEW), hosted on AWS
- Amazon Bedrock integration using Anthropic Claude Instant (classification) and Claude 2.0 (test case generation)
- Four-step workflow: requirement upload → AI classification → human validation → AI test case generation with human review
- Human-in-the-loop gates at both classification and test case generation stages
- Architecture: Amazon API Gateway, AWS Lambda, Amazon DynamoDB for persistence
- CSV-based import/export for integration with existing requirement management and testing tools
- Implemented in production within four weeks
Results
The AI-assisted workflow reduced test case creation time by up to 80% , according to AWS's published blog. The human-in-the-loop approach is positioned as the mechanism for maintaining accuracy. The solution was in production within four weeks. The authors note that future enhancements could include direct ALM-to-testing API integrations and fine-tuning of the Bedrock models on accepted test cases to further improve quality.
Lessons
- Framing the AI as a draft generator that humans review and accept — rather than a fully autonomous system — is key to achieving both efficiency gains and quality assurance
- Role-specific prompt engineering (e.g., instructing the model to behave as a test engineer) and structured output formatting materially improve generation quality
- Separating classification from test-case generation into two sequential human-validated steps improves output reliability
- AWS Bedrock's multi-model API enables cost-optimized model selection per task (lighter model for classification, more capable model for generation)
- Implementing CSV import/export first allows rapid production deployment even before deeper ALM integrations are built