MSc in Reliability of LLM-Generated Code

Large Language Models (LLMs) are increasingly used to generate code and assist software development. While they can significantly boost productivity, LLM-generated code often contains subtle bugs, security vulnerabilities, and violations of engineering requirements. Ensuring the reliability and trustworthiness of such code is an open challenge at the intersection of software engineering, programming languages, and AI.

Throughout this MSc program, you will participate, as a member of Programming Languages Lab at Peking University, in research projects that investigate methods for assessing and improving the reliability of LLM-generated code:

• Developing techniques to check LLM-generated code against formal or semi-formal specifications, using program analysis, logic, and verification methods (e.g. static analysis, model checking, or contract checking).
• Designing tools and workflows that help developers and researchers detect, explain, and repair defects in LLM-generated code, with a particular focus on research software.
• Studying engineering requirements for integrating LLM-based tools into real development pipelines, including traceability, testing, reproducibility, and documentation.
• Evaluating these techniques in collaboration with industry partners and research groups to understand how they can best support realistic software and research projects.

Applicants should:

• Pass Peking University's postgraduate entrance examination.
• Have solid programming skills (e.g. in Python, Rust, OCaml).
• Have good English communication skills.
• Have the ability to work and think creatively both independently and in a team.
• Have a strong interest in software engineering for reliability, research software, and developer tooling.
• Ideally, have some background or interest in one or more of: program analysis, logic and verification, or AI/ML for code.
• Ideally, have previous research experience at the undergraduate level.

Interested applicants should send their CV and transcripts to Dr. Sergey Mechtaev at mechtaev@pku.edu.cn.