Maximilian Böther

prof_pic.jpg

I am a final-year Ph.D. student in Computer Science at ETH Zurich’s Systems Group and the Efficient Architectures and Systems Lab (EASL), supervised by Ana Klimovic and Gustavo Alonso. I am also a Member of Technical Staff at DatologyAI.

I work on the intersection of systems and data-centric AI. I am excited about data loaders, the data and systems aspects of large-scale LLM/VLM training, data management for machine learning, and distribution shift in real-world machine learning pipelines.

I have published open-source projects like Mixtera (Github), a lightweight data plane for LLM/VLM training, and Modyn (Github), a platform for training models on datasets that grow over time. I also contributed to the training of Apertus v1, Switzerland’s national LLM.

I published at venues such as SIGMOD, VLDB, MLSys, ACL, and ICLR, and interned at Google and Apple. I obtained B.Sc. and M.Sc. degrees in IT-Systems Engineering from Hasso Plattner Institute, Potsdam, Germany in 2020 and 2022. Please find my CV here.

news

Jun 5, 2026 I attended SIGMOD’26 in Bengalore, India, presenting Mixtera both at the main conference as well as giving an invited talk on it at the DEEM workshop.
May 12, 2026 We just released a report on how data curation alone can increase VLM quality across 20 public benchmarks.
Apr 30, 2026 I attended EuroSys’26 in Edinburgh and presented the current status of the data loader that we are building at DatologyAI. I also gave an Invited Talk at GreenSys’26 on efficient data mixing and loading for foundation model training.
Apr 10, 2026 We just released a cost-performance trade-off study for test-time agent adaptation that has been accepted to the LLA workshop at ICLR‘26!
Apr 7, 2026 Our paper on Apertus v1, Switzerland’s national LLM, has been accepted to ACL‘26!
Mar 17, 2026 Ever wondered if you should include your domain-specific finetuning data in your pretraining mix? Check out The Finetuner’s Fallacy, where we dig into this question!
Feb 20, 2026 After finishing my internship at DatologyAI, I am happy to continue as Member of Technical Staff, working on our (soon-to-be) open source data loader!
Feb 16, 2026 Check out ÜberWeb, where we present our insights on curating multilingual data at the 20 trillion token scale.
Jan 6, 2026 I gave a talk at the DEEM Lab Research Seminar on Data Engineering for ML at TU Berlin, hosted by Sebastian Schelter.
Dec 7, 2025 I attended NeurIPS’25 in San Diego.
Nov 23, 2025 Mixtera has just been accepted to SIGMOD’26 in Bengaluru, India!
Oct 2, 2025 We organized the Building Efficient System Infrastructure for AI track at the AI+X Summit 2025 in Zurich and I gave a presentation about Mixtera.
Sep 15, 2025 I joined DatologyAI as a Research Intern! I will be working on data loading for LLM/VLM training.
Sep 1, 2025 We just released Apertus v1, Switzerland’s national LLM. Happy to have contributed to the pretraining data! The models can be found on huggingface and the technical report on Github.
Jun 27, 2025 I attended SIGMOD’25 in Berlin and presented Modyn at the conference!
May 19, 2025 I started as an ML Intern at Apple  in Seattle, working on reinforcement learning infrastructure for diffusion models!
May 15, 2025 I presented our distributed data selection paper at MLSys 2025, and gave a talk at DatologyAI on Mixtera.
Mar 19, 2025 I have received the ML and Systems Rising Star Award 2025. Thank you so much!
Mar 4, 2025 I presented Mixtera and Modyn at BTW’25. Thank you for the great discussions!
Mar 1, 2025 We just released a preprint on Mixtera, our data plane for foundation model training. If you are training LLMs or VLMs, and are looking for infrastructure for data loading and mixing, please feel free to reach out!
Feb 10, 2025 Our paper on distributed submodular subset selection–a result from my Google internship–has been accepted to MLSys’25. See you soon in Santa Clara!
Oct 31, 2024 Our paper on Modyn has been accepted to SIGMOD’25 in Berlin!
Oct 4, 2024 We organized the Systems for Cost-Efficient AI Track at the AI+X Summit in Zurich.
Sep 30, 2024 Our vision paper on Mixtera, our lightweight data lake for LLM training, has been accepted to HotInfra’24 at SOSP. See you in Austin, TX!
Aug 2, 2024 Happy to have attended the Dagstuhl Seminar 24311: Resource-Efficient Machine Learning.
Jun 17, 2024 I will talk about Modyn at the Data-centric Machine Learning (DML) workshop at ICLR’24. See you in Vienna!
Feb 26, 2024 We just released a preprint of our paper on scaling out practical subset selection using submodular functions. This paper is a result of my internship at Google.
Jun 17, 2023 Our paper on analyzing vectorized hash tables across CPU architectures just got accepted at VLDB’23 in Vancouver!
Jun 5, 2023 I joined Google for a summer research internship in Sunnyvale, California, USA! I am working on scaling out submodular data subset selection.
Apr 11, 2023 Our work-in-progress workshop paper on Modyn, our research platform for model training on dynamic datasets, has been accepted at EuroMLSys’23 in Rome!
Nov 1, 2022 I joined the ETH Zurich Systems Group and the Efficient Architectures and Systems Lab (EASL) to do a Ph.D. in Machine Learning Systems, supervised by Professor Ana Klimovic. Looking forward to the new adventures in Switzerland!
Oct 15, 2022 Our paper on efficiently computing directed minimum spanning trees (arboresence) has been accepted for publication at ALENEX 2023. Check out the final version here.
Jun 6, 2022 Our Law Smells paper, which applies concepts of software engineering to the law, has been published in AI&Law. Check out the final version here.
Jan 24, 2022 Our paper on deep learning for combinatorial optimization just got accepted at ICLR! Check out the final version here.
Dec 27, 2021 This website just went online!

selected publications

  1. ACL
    Apertus: Democratizing Open and Compliant LLMs for Global Language Environments
    Hernández-Cano, Alejandro, Hägele, Alexander,  Huang, Allen Hao and 99 more authors
    In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL) 2026
  2. SIGMOD
    Mixtera: A Data Plane for Foundation Model Training
    Böther, Maximilian, Yao, Xiaozhe,  Kerimoglu, Tolga and 3 more authors
    In Proceedings of the Conference on Management of Data (SIGMOD) 2026
  3. MLSys
    On Distributed Larger-Than-Memory Subset Selection With Pairwise Submodular Functions
    Böther, Maximilian, Sebastian, Abraham,  Awasthi, Pranjal and 2 more authors
    In Proceedings of the Conference on Machine Learning and Systems (MLSys) 2025
  4. SIGMOD
    Modyn: Data-Centric Machine Learning Pipeline Orchestration
    Böther, Maximilian, Robroek, Ties,  Gsteiger, Viktor and 3 more authors
    In Proceedings of the Conference on Management of Data (SIGMOD) 2025
  5. VLDB
    Analyzing Vectorized Hash Maps Across CPU Architectures
    Böther, Maximilian, Benson, Lawrence,  Klimovic, Ana and 1 more author
    Proceedings of the VLDB Endowment 2023