Building large-scale AI agents and ML solutions at BMW Group

This page provides deeper technical context on selected AI, Generative AI, and cloud projects I’ve led and contributed to.

About me

I am an engineering leader driving cloud platforms and large-scale AI and Generative AI systems. I have over two decades of industry experience, including more than a decade leading cross-functional teams on high-impact AI, cloud, and strategic initiatives. At the BMW Group, the machine learning teams I built and led developed features and models interacting with millions of vehicles.

My career has evolved from writing real-time embedded C/C++ for the BMW Superbike racing team to architecting enterprise-scale Generative AI solutions. Currently, I lead teams building LLM-based agents at the BMW Group, using GenAI to optimize productivity for hundreds of software developers as well as the Connected Vehicle backend serving 23M vehicles. I combine hands-on system design with end-to-end technical ownership.

I also enjoy teaching an introductory hands-on course on Machine Learning, Generative AI, and LLM-based agents at Fresenius University Munich, which also covers Data Science and Reinforcement Learning for LLM and games (Course Materials & Notebooks).

Additionally, I currently hold 9 professional certifications across AWS, Google Cloud and agile leadership.

🏆 Featured Highlights

Strategic Pillar Foundational AI Reliability Enterprise Cloud Operations
Innovation Strategy Structural testing of LLM-based agents
One of the first frameworks enabling unit- and integration-level testing of LLM-based agents, adopted within a BMW GenAI platform.
🏆 BMW Connected Company Innovation Award (Q1 2025) for this work.
Publication: IEEE | Open access: arXiv | Source: GitHub		 GenAI cloud optimization bot
Optimizing cloud accounts at scale via LLM-based multi-agent system.
Case study: AWS Case Study | Video: AWS re:Invent Innovation Talk | Publication: AWS blog | Additional reach: Facebook post (~1.9M views)
Platform Engineering Generative AI toolkit
Lead architect for open-source LLM lifecycle automation.
Publication: arXiv | Source: GitHub		 Scaling end-to-end cloud governance
Automating compliance for a cloud backend for 23M vehicles with 1,300+ services.
Publication: AWS Blog

🛠 Selected Projects

Structural testing for LLM-based agents

2025 | Publication: IEEE | Open access: arXiv | Source code: GitHub Repo

Framework enabling unit- and integration-level testing of LLM-based agents beyond acceptance tests. Combines OpenTelemetry-based tracing, automated verification and component mocking (including LLM) to enable software engineering best practices (TDD, Regression testing) for GenAI. I ideated the overall approach and contributed to the core implementation.

Generative AI toolkit

2024 | Publication: arXiv, Source code: GitHub Repo

Open-source framework that automates core workflows across the lifecycle of LLM-based applications—configuration, testing, monitoring, and optimization. I ideated and co-designed the toolkit and served as technical owner for its development.

Cloud Optimization Bot

2023 | Publication: AWS Case Study, AWS blog, Video: AWS re:Invent Innovation Talk 2023, additional reach: Facebook (~1.9M views)

One of the earliest LLM-based multi-agent systems. Built on Amazon Bedrock to optimize cloud infrastructure, delivering eight-digit annual savings. I ideated and co-designed the system and served as the technical owner leading the development.

Scaling end-to-end cloud governance

2023 | Publication: AWS Blog, Video: AWS re:Invent 2023

Automated compliance framework enforcing internal standards across hundreds of AWS accounts at scale. I ideated the framework, co-led the project, and contributed directly to the development.

Proactive Care

2022 | Video: AWS re:Invent 2022

Led a cross-functional prototyping team delivering an end-to-end automotive diagnosis workflow at scale (1M vehicles). Later rolled out fleet-wide to 23M vehicles by BMW aftersales organization.

View more projects ...

Speeding up cloud incident root cause analysis

2025 | Publication: AWS Blog

Designed and built an LLM-based agent with Amazon Bedrock to automate incident root-cause analysis across distributed cloud services. Reduces diagnosis time from hours to minutes with ~85% accuracy. I ideated the solution and co-led its development.

Ticketron

2024 | Video: AWS re:Invent 2024

Streamlining defect ticket workflows using ML for duplicate detection and automated routing. Reduced manual triage effort and accelerated the software development lifecycle. I served as the technical lead for its development and integration.

Headunit in the cloud (vECU)

2023 | Publication: AWS Blog

Virtualization of automotive Android control units on cloud instances to scale testing for 20M+ vehicles. I led the responsible BMW team, providing strategic guidance and architectural oversight.

Cloud migration of BMW ConnectedDrive

2023 | Publication: AWS Blog

One of the largest and earliest successful migrations of an automotive OEM to the cloud. For two years, I led the team responsible for cloud architecture, architectural standards, and the migration initiative covering over 1,300 microservices.

Lecturer, Artificial Intelligence & Data Science

2020 – Present | Course Materials & Notebooks

Introductory hands-on Master’s course for economics students. Curriculum spans Data Science fundamentals, Classical Machine Learning, Generative AI for text and images, building LLM-based agents and exploring Reinforcement Learning for LLM and games.

Superbike World Championship Racing Team

2005

Developed real-time C++/RTEMS modules for high-frequency data logging (1kHz) in professional racing environments. I contributed as an embedded software developer.

Software BMW S85 Engine for BMW M5 (E60/E61)

2004

Developed a code-signing tool in C/C++ for the M5 ECU and designed a Hardware-in-the-Loop (HiL) test unit to validate electronic throttle control reliability. I was a tool developer in this team.


📜 Professional Certifications

I hold several certifications across machine learning, cloud development, and agile methodologies.

View all certifications

Amazon Web Services (AWS)

  • AWS Certified Machine Learning – Specialty
  • AWS Certified Data Analytics – Specialty
  • AWS Certified Data Engineer – Associate
  • AWS Certified Developer – Associate

Google Cloud (GCP)

  • Professional Data Engineer
  • Professional Machine Learning Engineer
  • Generative AI Leader

Agile & Leadership

  • Professional Scrum Master I (PSM I)
  • Professional Scrum Product Owner I (PSPO I)

All credentials can be verified on my Credly Profile.


📝 Publications

You can also check my Google Scholar account.

  1. J. Kohl, O. Kruse, Y. Mostafa, A. Luckow, K. Schroer, T. Riedl, R. French, D. Katz, M. Luitz, T. Takher, K. Friedl, C. Laurent-Winter: Automated structural testing of LLM-based agents: methods, framework, and case studies. Proceedings of the IEEE Big Data Conference 2025, pp. 1847-1856, 2025. IEEE | arXiv.
  2. I. Drave, B. Rumpe, A. Wortmann, J. Berroth, G. Hoepfner, G. Jacobs, K. Spuetz, T. Zerwas, C. Guist and J. Kohl: Modeling mechanical functional architectures in SysML. In Proceedings of the 23rd ACM/IEEE International Conference on Model Driven Engineering Languages and Systems, pages 79–89, 2020. Link.
  3. J. Wildgruber, J. Kohl, T. Bindel, L. Gloger, T. Takher, O. Kruse, H. Vu, A. Krisnan, S. Saxena, K. Robins: Innovating at speed: BMW’s Generative AI solution for cloud incident analysis, 2025. Link.
All publications

Peer-reviewed publications

  1. J. Kohl, O. Kruse, Y. Mostafa, A. Luckow, K. Schroer, T. Riedl, R. French, D. Katz, M. Luitz, T. Takher, K. Friedl, C. Laurent-Winter: Automated structural testing of LLM-based agents: methods, framework, and case studies. Proceedings of the IEEE Big Data Conference 2025, pp. 1847-1856, 2025. IEEE | arXiv.
  2. T. Zerwas, G. Jacobs, L. Brand, S. Dehn, K. Spütz, G. Höpfner, C. Matz, C. Guist, J. Berroth, C. Konrad, J. Kohl: Customer-centric and function-oriented development of mechatronic systems. In Forschung im Ingenieurwesen, 2023. Link.
  3. G. Höpfner, I. Nachmann, T. Zerwas, J. Berroth, J. Kohl, C. Guist, B. Rumpe, G. Jacobs: Towards a Holistic and Functional Model-Based Design Method for Mechatronic Cyber-Physical Systems. In Journal of Mechanical Design, 2023. Link.
  4. G. Höpfner, G. Jacobs, T. Zerwas, I. Drave, J. Berroth, C. Guist, B. Rumpe and J. Kohl: Model-based design workflows for cyber-physical systems applied to an electric-mechanical coolant pump. In IOP Conference Series: Materials Science and Engineering, volume 1097, page 012004. IOP Publishing, 2021. Link.
  5. T. Zerwas, G. Jacobs, K. Spütz, G. Höpfner, I. Drave, J. Berroth, C. Guist, C. Konrad, B. Rumpe and J. Kohl: Mechanical concept development using principle solution models. In IOP Conference Series: Materials Science and Engineering, volume 1097, page 012001. IOP Publishing, 2021. Link.
  6. I. Drave, B. Rumpe, A. Wortmann, J. Berroth, G. Hoepfner, G. Jacobs, K. Spuetz, T. Zerwas, C. Guist and J. Kohl: Modeling mechanical functional architectures in SysML. In Proceedings of the 23rd ACM/IEEE International Conference on Model Driven Engineering Languages and Systems, pages 79–89, 2020. Link.
  7. I. Krüger, M. Menarini, F. Seracini, M. Fuchs and J. Kohl: Improving the development process for automotive diagnostics. In 2012 International Conference on Software and System Process (ICSSP), pages 63–67. IEEE, 2012. Link.
  8. J. Kohl: Effiziente Diagnose von verteilten Funktionen automobiler Steuergeräte. PhD thesis, Technische Universität München, 2012. Link.
  9. J. Kohl, A. Kotucz, J. Prenninger, A. Dorneich and S. Meinzer: Using multivariate split analysis for an improved maintenance of automotive diagnosis functions. In 15th European Conference on Software Maintenance and Reengineering, pages 305–308. IEEE, 2011. Link.
  10. J. Kohl and A. Bauer: Role-based diagnosis for distributed vehicle functions. In Proceedings of the 21st International Workshop on the Principles of Diagnosis (DX), 2010. Link.

Non-peer reviewed publications

  1. M. Menzel, J. Kohl: Small models, high quality: Inside BMW Group’s experiments evaluating domain-specific language models, 2026. Google Blog.
  2. J. Wildgruber, J. Kohl, T. Bindel, L. Gloger, T. Takher, O. Kruse, H. Vu, A. Krisnan, S. Saxena, K. Robins: Innovating at speed: BMW’s Generative AI solution for cloud incident analysis, 2025. AWS Blog.
  3. P. Altmann, A. Gross, C. Müller, J. Kohl, J. Frielinghaus, L. Mödl, S. Marzani: How BMW is using AWS as native development target to scale and automate embedded software development, 2025. AWS Blog.
  4. J. Kohl, L. Gloger, R. Costa, O. Kruse, M. Luitz, D. Katz, G. Barbeito, M. Schweier, R. French, J. Schroeder, T. Riedl, R. Perri, Y. Mostafa: Generative AI Toolkit - a framework for increasing the quality of LLM-based applications over their whole life cycle, arXiv preprint arXiv:2412.14215, 2024. arXiv.
  5. S. Kallin, S. Molau, J. Kohl, B. Jones, C. Müller: How BMW Group migrated the BMW ConnectedDrive backend to AWS, 2024. AWS Blog.
  6. J. Kohl, B. Jones, C. Müller, D. Engelhardt, L. Zinovyeva, N. Castro, S. Saxena, S. Khodjaev: BMW Group develops a GenAI assistant to accelerate infrastructure optimization on AWS, 2024. AWS Blog.
  7. J. Kohl, B. Jones, C. Müller, D. Engelhardt, M. Arrais, S. Kallin: How BMW Group use automation to achieve end-to-end compliance at scale on AWS, 2024. AWS Blog.
  8. N. Olivain, P. Tiefenbacher and J. Kohl: Bayesian structural learning for an improved diagnosis of cyber-physical systems, arXiv preprint arXiv:2104.00987, 2021. arXiv.


🎤 Public Talks

  1. J. Kohl: Developing a GenAI assistant to accelerate cloud infrastructure optimization. AWS re:Invent 2023, 2023. Video.
  2. J. Kohl: Scaling end-to-end cloud governance. AWS re:Invent 2023, 2023. Video.
  3. J. Kohl: BMW Group’s AI-based automotive Proactive Care Workflow. AWS re:Invent 2022, 2022. Video.
Selected public talks
  1. J. Kohl: Optimizing automotive software development workflows with AI/ GenAI. Move MOBILITY RE-IMAGINED 2026, 2026.
  2. J. Kohl: Automated Structural/ White-Box Testing of LLM-Based Agents. EuroSTAR 2026, 2026.
  3. J. Kohl: Optimizing automotive software development workflows with AI/ GenAI. European AI and software-defined vehicle summit 2026, 2026.
  4. J. Kohl: Sports and artificial intelligence. Professional Darts Corporation (PDC) Europe Poker Charity Turnier 2026, 2026. Slides.
  5. J. Kohl: Automated structural testing of LLM-based agents. Google 3 Minutes of AI IRL, 2025.
  6. A. Krishnan, J. Kohl: Accelerated Root Cause Analysis at BMW with Amazon Bedrock Agents. AWS Deep Dive Days 2025, 2025.
  7. J. Kohl, Y. Mostafa, M. Radulescu Kober: Evaluating and testing customer-facing Generative AI systems at scale. AWS Deep Dive Days 2025, 2025.
  8. O. Schnaack, K. Robins, J. Kohl: BMW speeds car development with a new app for defect ticket routing, AWS re:Invent 2024, 2024. Video.
  9. J. Kohl, L. Gloger, O. Kruse: Increasing the quality of LLM-based agents by accelerating the feedback loop. AWS Deep Dive Days October 2024, 2024.
  10. J. Kohl, M. Tava: Using Generative AI To Optimize Cloud Rooms Performance and Costs. InfoQ Dev Summit, 2024.
  11. J. Kohl, C. Müller: How BMW scales the software-defined vehicle using cloud-native tools. AWS Summit 2024, 2024.
  12. J. Kohl, C. Müller: BMW Group accelerates cloud optimization using Amazon Bedrock. AWS Deep Dive Days 2024, 2024. Slides.
  13. J. Kohl: Developing GenAI assistant to accelerate cloud infrastructure optimization. AWS re:Invent 2023, 2023. Video.
  14. J. Kohl: Scaling end-to-end cloud governance. AWS re:Invent 2023, 2023. Video.
  15. J. Kohl: BMW Group’s AI-based automotive Proactive Care Workflow. AWS re:Invent 2022, 2022. Video.
  16. J. Kohl: Vision for Engineering Workplace 2025. 4th International IPS Cable Simulation Conference 2019, 2019.


📜 Patents

  1. A. Ganser, J. Kohl: Method and System for Migrating Software between cloud systems, 2025. Link: German patent library DPMA.
  2. P. Tiefenbacher, T. Billinger, J. Kohl: Method for operating a motor vehicle comprising an internal combustion engine, control unit and motor vehicle, 2021. Link: German patent library DPMA.
  3. C. Guist, J. Kohl: Method for providing a component using additive manufacturing, 2019. Link: German patent library DPMA.

Note: 14 additional AI and GenAI patents pending.