Research

Research for the Embedded Development of Tomorrow

My work focuses on applied research at the intersection of embedded software engineering, DevSecOps, and artificial intelligence. The goal: to make development processes not only faster but also more reliable and intelligent - with a strong emphasis on test automation and AI-assisted engineering support.

Publications & Conference Talks

I regularly publish in conference proceedings and present at national and international conferences - primarily on topics such as AI in embedded development, automated testing, and secure CI/CD processes.

Agent-Driven DevSecOps: Transforming Embedded Software Development

Embedded World Conference 2026: Paper, Presentation, Conference

How can current AI technologies improve the effectiveness of DevSecOps workflows? This talk presents a containerized multi-agent infrastructure based on open standards like ACP (Agent Client Protocol), MCP (Model Context Protocol), and Agent Skills, alongside a labgrid MCP server that enables AI-assisted CI/CD failure triage with controlled access to embedded devices. Practical experience with security isolation, audit trails, and a roadmap toward autonomous multi-agent collaboration via the A2A protocol are also covered.

Key Takeaways

• Containerized multi-agent infrastructure with ACP + MCP + Agent Skills and security isolation
• Labgrid MCP server for AI-assisted CI/CD failure triage with controlled device access
• CI failure triage reduced from ~15 minutes to ~2 minutes
• Roadmap toward collaborative multi-agent scenarios using the A2A protocol

Conclusion: AI agents relieve embedded engineers of repetitive analysis tasks and shorten feedback cycles. Quality in DevSecOps remains a choice: engineers must carefully judge what to delegate and critically evaluate AI outputs at every step.

AI-Driven DevSecOps for Embedded Systems

IT-SECX 2025: Presentation, Conference

This talk combines pytest/labgrid with emerging protocols such as MCP (Model Context Protocol) and ACP (Agent Communication Protocol). The result: a self-learning test and analysis ecosystem that automatically generates, executes, and continuously improves tests based on real-world feedback.

Key Takeaways

• AI-driven test generation and automated hardware orchestration
• Integration of MCP & ACP for reproducible multi-agent workflows
• First steps toward AI-based log file analysis and error classification
• Future vision: No-Code Security - enabling security validation without programming

Conclusion: AI-DevSecOps turns embedded testing into a self-learning, secure, and scalable system.

Qualification of AI for Embedded Systems Testing

Embedded World Conference 2025: Paper, Presentation, Conference

How much of test development can artificial intelligence take over without compromising quality? This paper demonstrates how Large Language Models (LLMs) such as GPT-4 or Llama 3 can automate up to 70 % of test preparation, seamlessly integrated into pytest/labgrid-based CI/CD pipelines.

Key Takeaways

• Automatic generation of test code from Markdown-based specifications
• Integration with GitLab, QEMU simulation, and physical hardware
• Improved traceability and IEC 62443 / CRA compliance
• Reduction of manual effort by up to 70 % - with consistent code quality

Conclusion: AI doesn’t replace engineers - it makes them three times faster and ensures security and compliance are built in from the start.

Embedded Security Testing

IT-SECX 2024: Presentation, Conference

Secure embedded devices require reproducible tests - but not necessarily complex frameworks. This talk shows how pytest and labgrid can be used to implement IEC 62443-compliant tests that remain simple, maintainable, and fully CI/CD-compatible.

Key Takeaways

• Security and functional testing inside a Docker Compose environment
• Example: OpenVPN client tests in just ~500 lines of total code
• Clear distinction between functional and non-functional requirements
• Executable on both QEMU and physical hardware

Conclusion: Security testing should be part of daily development - not a separate discipline.

Boosting Embedded System Development: A Case for Rapid Testing

Embedded World Conference 2024: Paper, Presentation, Conference

Lengthy integration tests slow down developers and delay releases. This talk presents how Hardware-in-the-Loop (HiL) tests can be accelerated algorithmically using pytest and labgrid - without compromising reliability or reproducibility.

Key Takeaways

• Up to 85 % shorter test cycles through parallel execution on multiple DUTs
• Algorithm based on the Cutting-Stock Problem for optimal test distribution
• Fully compatible with official pytest APIs → instantly CI/CD-ready
• Example: 4.5-hour test suite reduced to 35 minutes using 20 DUTs

Conclusion: Parallel testing isn’t a luxury - it’s the key to faster feedback, higher productivity, and better quality.