Master Thesis: Design, Prototype Development and Evaluation of an On-Premise AI Assistant System for Test Stand Engineering

As a leading automobile manufacturer, Dr. Ing. h.c. F. Porsche AG is revolutionising test stand engineering with the help of generative AI.

The goal of this master thesis is to design, prototype and evaluate a fully on-premise AI assistant system based on Retrieval-Augmented Generation (RAG).

The successful candidate will:

• Conduct literature research on locally operable language models (LLMs) and RAG architectures
• Design a fully local AI architecture based on existing hardware
• Set up a RAG pipeline to access the manufacturer's documentation of the automation system
• Provide the system via a web-based user interface on the departmental network
• Extend the approach to retrieve and adapt tested test stand parameter sets from past projects
• Evaluate the system's response quality and determine its limitations using a catalogue of questions developed with experts
• Prepare a presentable demonstrator, document the architecture and derive recommendations for further expansion

Requirements:

• Study programme: Mechanical Engineering, Mechatronics, Computer Science, Data Science, Electrical Engineering or similar
• Study focus: Artificial Intelligence / Machine Learning, Software Development, Data Analysis and Data Preparation, Automation and Test Stand Technology
• Technical knowledge: Basic understanding of language models (LLMs) and Natural Language Processing, interest in designing and building AI and retrieval systems
• IT skills: Python, basic knowledge of Linux / WSL, experience with databases and containerisation (Docker) beneficial
• Language skills: English or German

Benefits:

• Opportunities for professional growth and development
• Collaborative and dynamic work environment
• Flexible working hours and work-life balance

How to apply:

Please submit your application via our online application form, including your CV, cover letter, academic transcripts and any additional relevant documents.