AUMOVIO improves quality of automotive software at scale using multi-agent AI on Amazon Bedrock

This article describes how AUMOVIO developed a multi-agent AI system called the problem finder that detects automotive software issues missed by traditional static analysis tools.

• Uses problem templates grounded in historical issue data to guide AI agents through systematic code analysis
• Combines cross-component specifications, AUTOSAR configurations, and implementation code in single analysis context
• Supports full analysis and delta analysis modes for comprehensive audits and continuous integration workflows
• Detects semantic inconsistencies like signal routing failures and value mapping inversions before vehicle integration
• Includes human-in-the-loop feedback cycle where domain experts refine templates based on findings
• Runs on Amazon Bedrock AgentCore with multi-agent architecture separating orchestration and analysis concerns
• Found 30+ high-priority issues in production component that passed all traditional tooling

The problem finder closes the gap between development and late-stage testing by catching integration issues early, reducing costly rework in automotive software development.