Master's thesis: AI agents & simulation development

Model-based simulation development is time-consuming and requires extensive domain knowledge. In this six-month master's thesis at Bosch Rexroth, you will investigate how coordinated AI agents can speed up this process. The focus is on agentic engineering, developing controllable agent-based workflows and using local AI models.

• Analyze patterns, anti-patterns and typical obstacles when building agent-based workflows, especially to prevent and detect hallucinations
• Develop a harness engineering approach based on LangChain and LangGraph for structured control of processes, tool access and checking steps
• Investigate the transition from cloud-based to local AI models, including pre-training and fine-tuning
• Evaluate performance in terms of token usage and computing power, and secure result quality through systematic evals and defined test cases
• Investigate from a UX perspective how developed simulations can be intuitively moved into productive use with or without an AI agent
• Develop a practical, controllable and hand-over-ready approach for AI-supported simulation development

• Education: You are completing a master's degree in computer science, artificial intelligence / data science, computational engineering, engineering informatics or a comparable technical program.
• Programming and AI: You have very good Python programming skills and solid practical experience with LangChain and LangGraph (state management, routing, tool integration).
• LLM know-how: You bring knowledge of prompt engineering patterns, structured outputs and methods for preventing and detecting hallucinations, as well as working with open-source LLMs (e.g. Llama, Mistral), fine-tuning (e.g. PEFT/LoRA) and model adaptation.
• Ideally basic knowledge of model-based simulation and engineering workflows (e.g. Matlab/Simulink, Modelica, FMU)
• Ideally first experience with systematic LLM evals and benchmarking of token efficiency and hardware resources (inference latency, VRAM)
• Ideally interest in human-AI interaction and interface usability for productive use
• Ideally experience with Git, Docker and inference engines for local models (e.g. vLLM, Ollama)
• You work analytically and solution-oriented and break down complex engineering problems systematically to model them in software
• You work in a structured and quality-conscious way with high standards for robust software architecture, error handling and scientific validation (harness engineering)
• You work independently and dive intensively into current AI developments
• Languages: You have fluent German or very good English skills, spoken and written.