Master's thesis on the use of generative AI in requirements engineering

Context

Modern vehicles have well over 100 control units and several hundred functions. This results in a high level of complexity for the development of these vehicles. Customer requirements for the entire vehicle must be broken down and detailed for the subsystems and functions. For each of the functions and control units, there is ultimately a large number of requirements, which are available in a semi-structured form in a combination of formal language and text, and a system specification derived from this. In addition, legal framework conditions, for example from standards and legal texts, must be taken into account.

This results in a five-digit number of requirements for a vehicle development project, which poses challenges for practical work, e.g. contradictions or inconsistencies in the requirements and specifications.

The master's thesis examines the extent to which Large Language Models (LLMs) can be used to support the development process. For example, requirements catalogs can be examined for inconsistencies or test cases can be extracted from requirements.

Tasks

• Investigation of approaches from the field of LLMs for analyzing requirements
• Selection of suitable models (e.g. open source models, ChatGPT, Google Gemini etc.)
• Investigation of the suitability of adaptation methods (prompt engineering, various forms of fine-tuning)
• Development of methods to quantify the handling of known problems of LLMs (e.g. hallucinations)
• Implementation of an exemplary use case with selected model and exemplary catalog of requirements
• Validation of the performance based on the implemented use case

Requirements

• You have programming skills (e.g. in Python).
• You have experience in the field of AI, data science or LLMs
• You have very good written and spoken German and English skills