Lead AI Engineer / Architect - Agentic Systems (m/f/d)

In our laboratory group, thousands of laboratory parameters are measured daily, findings are released and request certificates are processed. These processes are carried out manually by highly qualified MTAs and specialists. Our newly established AI unit, directly connected to the management, develops productive ML systems that digitize these core processes: from the computer-aided finding of medical samples through automated order acquisition to the cross-site scaling of these solutions to over 30 laboratories. The application is carried out in an ISO-15189 certified environment under the requirements of the GDPR, IVDR and the EU AI Act. Our goal is to answer solutions from the first pilot to the stable operation across the entire laboratory group. In your role, you are responsible for architecture and promote the following topics together with our team: Multi-Agent Systems & Tech Scouting: Architectural transformation and scaling of our internal agent ecosystem. Through proactive technology scouting in the open source market (e.g. evaluation of LangGraph updates), you are driving the continuous improvement of our development cycles in agile sprints. Hybrid LLM infrastructure: evaluation, linking and productive use of local vLLM inference clusters and commercial API models. You design efficient architectures for review and planning tasks, document interpretation and internal knowledge queries System integration: Development of MCP servers as a standardized interface for our agent ecosystems. This includes the deterministic provision of tools, prompts and context data for secure inference by our LLMs Clinical AI & group-wide ML solutions: Development of productive ML models as a central component of our value creation. In addition to our current focus on the medical core area, you will also implement accurate ML architectures for operational and commercial processes throughout the company ML pipelines, monitoring & retraining: build and operate robust data and ML pipelines. You make sure that our productively used models are performing at the various laboratory locations, are continuously monitored for data drift and are subjected to retraining in a quality-proof manner. This also includes the asynchronous evaluation of our models Composite scaling: Transfer of proven, value-generating ML architectures and AI systems from individual locations to the entire laboratory network, taking into account technical, regulatory and organisational requirements

Model Agnostics & System Architecture: You evaluate and select the optimal architectures for our various use cases. Based on sound trade-off analyses (cost, latency, quality), you design deterministic routing logics – be it the orchestration of commercial API LLMs, the use of local open-view models or the use of classic ML methods for structured laboratory data Experience in ''Agentic Engineering': You have not only chatted with LLMs, but have implemented complex workflows, state-machines or multi-agent systems programmatically (ideally with LangChain / LangGraph or in pure Python) and know how to keep 'Agentic Loops' stable and predictable in production Advanced Tech-Stack & local LLM hosting: Excellent Python skills and safe handling of the ML ecosystem (HuggingFace, PyTorch o.a.). You can also master structured prompt engineering as well as the provision of models outside of cloud APIs (vLLM, TGI) and apply concepts such as KV cache, quantization and tensor parallelism to multi-GPU setups safely End-to-End-MLOps & automated evaluation: You rolled out ML models and then operated, including error analysis, retraining and further development after the Go level. You use tracking and registration frameworks (e.g. MLflow, Weights & Biases) as well as model compilers (e.g. NVIDIA TensorRT) for operation and implement asynchronous test pipelines as well as automated benchmarking (e.g. LLM-as-a-Judge) for error analysis and quality assurance Practice Focus: Detectable production experience in Computer Vision, NLP / document processing or LLM integration Communication strength: You have the ability to understand, use-oriented, and to represent complex technical system design decisions vis-à-vis specialist physicians and management. Language knowledge: Flowing German (min. C1) for internal communication and safe technical English

You will find an attractive and modern workplace with us – including direct access to a dedicated Bare Metal GPU cluster in the local data center for unhindered, IVDR-compliant prototyping Direct, interdisciplinary cooperation at eye level with the medical management (Product Owner) during the expansion and scaling of our ML architecture Regular technical and organizational development opportunities are available to you through our Limbach Academy We offer you flexible working hours with the possibility of mobile work An unlimited working relationship is of course for us Thanks to our good transport connection you can easily reach us with our cheap Germany ticket or service bike lease We offer you an occupational pension including employer grant You will find your own canteen and free drinks at our house Regular corporate events and team events create a pleasant team atmosphere