Thesis project at Traton is an excellent way of making contacts for your future working life. Many of our current employees started their career with a thesis project.
Modern truck and heavy-vehicle platforms developed within the Traton Group rely on large volumes of embedded C code that must comply with strict safety, performance, and coding standard requirements. As software content increases, so does the engineering cost of writing, reviewing, and validating code manually. Large Language Models (LLMs) offer the possibility to automatically generate software artifacts, reducing development time and relieving engineers from repetitive implementation tasks.
In an industrial setting code quality cannot be compromised. Automatically generated code often fails to meet functional requirements or violates internal development guidelines such as MISRA C, or safety coding practices. For automotive software to be certifiable and deployable, such deviations must be detected and corrected efficiently.
To enable safe adoption of LLM generated code in an industrial workflow, it is essential to connect automated code generation with automated quality assurance. Verification tools can identify defects, but the key challenge is how to use their feedback to automatically repair the faulty code without human intervention. Different feedback strategies are possible, such as providing failing test cases, verification output, or guideline violations. It is not yet clear which strategy yields the most reliable repairs in practice, especially when multiple defects need to be fixed in a single iteration to reduce turnaround time and engineering effort.
Target:
The goal of this project is to explore methods for automatically repairing faulty LLM-generated programs using verification feedback. The focus is on designing and evaluating feedback strategies that improve LLM-generated code with respect to both functional and non-functional requirements. By experimenting with different types of feedback, the project aims to identify effective techniques for repairing programs and to analyze how well multiple issues can be fixed at once.
Example of assignments:
Education:
Education: MSc in Computer Science or similar, preferably (but not necessary) with some background in formal methods.
Number of students: 1-2.
Start date: January/February 2026.
Estimated time needed: 20-25 weeks.
Merlijn Sevenhuijsen, Industrial Ph.D., KTH / Traton, merlijn.sevenhuijsen@scania.com
Enclose CV, personal letter and transcript of records.
A background check might be conducted for this position.
Read Full Description