Scholar
Niels Mündler
Google Scholar ID: iX8Ib9wAAAAJ
PhD Student, ETH Zurich
Citations & Impact
Citations
529
H-index
6
i10-index
5
Publications
9
Co-authors
13
list available
Publications
Resume (English only)
Academic Achievements
- - 2025: Constrained Decoding of Diffusion LLMs with Context-Free Grammars
- - 2025: BaxBench: Can LLMs Generate Secure and Correct Backends?
- - 2025: Black-Box Adversarial Attacks on LLM-Based Code Completion
- - 2025: Type-Constrained Code Generation with Language Models
- - 2024: SWT-Bench: Testing and Validating Real-World Bug-Fixes with Code Agents
- - 2024: Self-contradictory Hallucinations of Large Language Models: Evaluation, Detection and Mitigation
- - 2021: Best Paper Award at ICTAC for 'A Verified Imperative Implementation of B+-Trees in Isabelle'
- - 2016: Second place at German national informatics contest (BWINF)
Research Experience
- - ETH Zurich, Research Associate at SRI Lab, January - June 2024
- - OpenSwap Technologies AG, Co-Founder and CTO, 2021 - 2023
Education
- - ETH Zürich, M.Sc. in Computer Science, October 2021 - December 2023
- - National University of Singapore, Exchange Student at School of Computing, August 2019 - January 2020
- - TU Munich, B.Sc. in Computer Science, October 2017 - March 2021
Background
- - Research Interests: Enhancing the reliability and security of Large Language Models, with an emphasis on ensuring the faithfulness and correctness of generated text and code
- - Professional Field: Programming languages and formal methods
- - Background: Since July 2024, pursuing a Ph.D. at the Secure, Reliable, and Intelligent Systems Lab at ETH Zürich, advised by Prof. Martin Vechev. Previously worked on the formal verification of data structures and algorithms and co-founded a start-up specializing in Smart Contracts.
Miscellany
- - Looking for motivated students with a strong background in programming languages and formal methods to conduct research projects, BSc or MSc theses
- - Interested in new project ideas around formal methods for Large Language Models, including extending type-constrained code generation (Go, Java, etc.) and improving the efficiency of constrained decoding for diffusion LLMs
