AUMOVIO Boosts Software Development with an Agentic Coding Assistant Powered by Amazon Bedrock

This article describes how AUMOVIO developed an AI-powered coding assistant for automotive software development using AWS services and Amazon Bedrock.

• AUMOVIO built agentic coding assistant to generate automotive code and unit tests
• Initial RAG approach failed; switched to agentic architecture with multi-step reasoning
• Solution integrates Cline and Amazon Q Developer with fine-tuned models via MCP protocol
• Fine-tuned Qwen3-32B model on 7,000 automotive functions using LoRA and SageMaker
• Architecture uses orchestrator models (Claude Sonnet) delegating to specialized fine-tuned models
• Evaluation shows fine-tuned models achieve superior C++ compliance and automotive standards adherence
• Solution reduced development time significantly; bug fixes from 5 days to minutes
• Future roadmap includes unit test generation and multi-agent framework expansion

AUMOVIO's specialized coding assistant demonstrates how domain-specific fine-tuning and agentic AI architectures can accelerate automotive software development while maintaining industry compliance standards.