# Tobias Lindenbauer

> Research Engineer at JetBrains working on post-training models for agentic applications in software engineering. This site hosts my research publications, selected projects, blog, and contact details. Based in Munich, Germany.

I am a Research Engineer at JetBrains working on post-training models for agentic applications in software engineering (for code and beyond)—agents that solve tasks over long horizons—with an emphasis on making them efficient and effective in terms of context management, memory, and compute. I completed my MSc in Computer Science at the Technical University of Munich (with distinction, 2025).

## Main pages

- [About](https://tobias.lindenbauer.me/): Background, research interests, and current work on post-training models for agentic applications in software engineering.
- [Publications](https://tobias.lindenbauer.me/publications/): Peer-reviewed and workshop publications on software engineering agents.
- [Projects](https://tobias.lindenbauer.me/projects/): Selected research and engineering projects, including a RouteLLM reproduction.
- [Blog](https://tobias.lindenbauer.me/blog/): Writing on machine learning, agents, and software engineering.

## Publications

- [The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Management](https://tobias.lindenbauer.me/publication/2025-12-01-complexity-trap): DL4C Workshop at NeurIPS 2025. Sliding-window observation masking matches LLM-based summarization for agent context management at a fraction of the compute cost.
- [GitGoodBench: A Novel Benchmark for Evaluating Agentic Performance on Git](https://tobias.lindenbauer.me/publication/2025-08-01-gitgoodbench): REALM Workshop at ACL 2025 (Spotlight). A benchmark for evaluating AI agent performance on version control tasks across three core Git scenarios.
- [From Knowledge to Noise: CTIM-Rover and the Pitfalls of Episodic Memory in Software Engineering Agents](https://tobias.lindenbauer.me/publication/2025-08-02-ctim-rover): REALM Workshop at ACL 2025. Repository-level understanding is pivotal for patch localization, and naive episodic-memory augmentation can introduce noise.

## Optional

- [Google Scholar](https://scholar.google.com/citations?hl=en&user=ms-Y9OgAAAAJ): Full list of articles and citations.
- [GitHub](https://github.com/LiQS-v2): Open-source code and projects.
- [LinkedIn](https://www.linkedin.com/in/tobias-lindenbauer): Professional profile.
- [Impressum](https://tobias.lindenbauer.me/impressum/): Legal notice and contact details (German TMG).