Master thesis: LLM Reasoning for Software Correctness

Join our Team

About the Opportunity

Would you like to contribute to the evolution of Ericsson’s 5G, 6G, and Cloud RAN solutions while exploring Artificial Intelligence and Software Engineering? This thesis opportunity focuses on improving software quality in large-scale systems where correctness, reliability, and availability are essential.

Modern development uses static and dynamic analysis, testing, code coverage, maintainability metrics, and defect history to identify quality issues. However, developers often need to manually interpret and correlate findings from multiple sources before deciding where to focus verification efforts.

Large Language Models (LLMs) offer strong capabilities in code understanding, reasoning, and defect analysis. Their use in large software systems is challenged by limited context windows, computational cost, and the risk of unsupported conclusions. This thesis will investigate whether LLMs can improve fault discovery by reasoning over structured software-quality evidence rather than raw source code alone.

The central hypothesis is that combining deterministic analysis with probabilistic LLM reasoning can identify additional fault candidates, improve verification prioritization, and reduce the cost of applying LLMs to large codebases.

What You Will Do

You will investigate and develop an evidence-guided approach to AI-assisted software quality analysis. Your work may include:

Collecting and structuring software-quality evidence from static and dynamic analysis, test coverage, maintainability metrics, code smells, historical defects, and bug reports.

Investigating whether analysis findings improve LLM-based defect detection and whether LLMs can identify fault candidates beyond those reported by existing tools.

Exploring compact, machine-consumable, and token-efficient representations of large codebases and quality evidence.

Evaluating which quality indicators provide the greatest analytical value relative to token usage and computational cost.

Analyzing how undefined behavior affects the reliability and trustworthiness of LLM-generated recommendations.

Investigating whether LLMs can help developers understand, explain, and prioritize potential false positives from static analysis.

Designing and conducting experiments to evaluate accuracy, effectiveness, and cost efficiency.

The Skills You Bring

Currently pursuing a master’s degree in Computer Science, Software Engineering, Data Science, Artificial Intelligence, Theoretical Computer Science, or a related field.

Strong programming skills in Python, C, or C++.

An interest in LLMs, Generative AI, software quality, and program analysis.

Knowledge of software testing, static or dynamic analysis, compilers, or software verification is beneficial.

Familiarity with machine learning, data science, or AI-assisted software development is a plus.

Strong analytical, problem-solving, communication, and presentation skills in English.

A curious, and self-driven mindset, with the ability to work independently and collaborate effectively.