Maximilian Böther

I am a Member of Technical Staff at DatologyAI. I work at the intersection of systems and data-centric AI, and am excited about the data and systems aspects of large-scale model training. In particular, I currently focus on data loading and processing for LLMs and VLMs.

After finishing my Ph.D., I am staying full time with DatologyAI to help build up our European operations. As part of this, I will be spinning up a training infrastructure team. If you are interested in working on these topics, please reach out!

I work on Zephon, our open-source, elastically deterministic data loader for text and multimodal training (paper). My earlier open-source projects include 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 completed my Ph.D. in Computer Science at ETH Zurich’s Systems Group and the Efficient Architectures and Systems Lab (EASL), supervised by Ana Klimovic and Gustavo Alonso. My work has been published at venues such as SIGMOD, VLDB, MLSys, NeurIPS, ICLR, ACL, and COLM, and I 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

Oct 7, 2026	Super excited that we are open-sourcing Zephon today! Zephon is an elastically deterministic data loader for text and multimodal training. Check out the project and our paper!
Oct 1, 2026	After finishing my Ph.D., I am excited to stay full time with DatologyAI and help build up our European operations! I will also be spinning up a training infrastructure team. If you are interested, please reach out!
Sep 25, 2026	DataComp-VLM: Improved Open Datasets for Vision-Language Models has been accepted to NeurIPS’26! Check out the latest entry in the DataComp benchmark series.
Sep 10, 2026	I am happy to serve on the program committee of MLSys’27!
Aug 24, 2026	I successfully defended my Ph.D. at ETH Zurich! My thesis is titled System Support for Declarative and Data-Centric Machine Learning. Many thanks to my committee members Martin Jaggi, Sebastian Schelter, and Steven Hand, and of course to my supervisor Ana Klimovic for her guidance and support throughout my Ph.D.!
Jul 8, 2026	The Finetuner’s Fallacy has been accepted to COLM’26! We study when to include domain-specific finetuning data in the pretraining mix.
Jun 5, 2026	I attended SIGMOD’26 in Bangalore, 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.

older news

selected publications

arXiv

Zephon: Elastic Determinism for Online, Stateful Foundation Model Data Loading Pipelines

Böther, Maximilian, Wills, Josh, Robroek, Ties and 14 more authors

arXiv preprint arXiv:2610.03087 2026 — Under revision at VLDB 2027

Bib HTML PDF Website

@article{Bother2026Zephon,
  author = {Böther, Maximilian and Wills, Josh and Robroek, Ties and Xu, Sonnet and Burstein, Paul and Zayas, Daniel and Blakeney, Cody and Joshi, Siddharth and Yin, Haoli and Adiga, Rishabh and Mongstad, Haakon and Merrick, Luke and Maini, Pratyush and Morcos, Ari and Leavitt, Matthew and Klimovic, Ana and Gaza, Bogdan},
  title = {Zephon: Elastic Determinism for Online, Stateful Foundation Model Data Loading Pipelines},
  journal = {arXiv preprint arXiv:2610.03087},
  year = {2026},
  eprinttype = {arXiv},
  eprint = {2610.03087},
  url = {https://arxiv.org/abs/2610.03087},
  note = {Under revision at VLDB 2027},
}

NeurIPS

DataComp-VLM: Improved Open Datasets for Vision-Language Models

Farina, Matteo, Udandarao, Vishaal, Nguyen, Thao and 33 more authors

In Advances in Neural Information Processing Systems (NeurIPS) 2026

Bib HTML PDF

@inproceedings{Farina2026DataCompVLM,
  author = {Farina, Matteo and Udandarao, Vishaal and Nguyen, Thao and Kuzucu, Selim and Böther, Maximilian and Hochlehnert, Andreas and Ghosh, Adhiraj and Nezhurina, Marianna and Roth, Karsten and Struber, Joschka and Zhang, Yuhui and Dziadzio, Sebastian and Sui, Elaine and Jahagirdar, Soumya and Ghosh, Dhruba and Hammoud, Hasan and De Min, Thomas and Caldarella, Simone and Mirza, Jehanzeb and Keh, Sedrick and Cherti, Mehdi and Kuehne, Hilde and Schiele, Bernt and Yeung-Levy, Serena and Naeem, Muhammad Ferjad and Tombari, Federico and Klimovic, Ana and Ricci, Elisa and Bethge, Matthias and Oh, Sewoong and Prabhu, Ameya and Tonioni, Alessio and Jitsev, Jenia and Mancini, Massimiliano and Schmidt, Ludwig and Parthasarathy, Nikhil},
  title = {DataComp-VLM: Improved Open Datasets for Vision-Language Models},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2026},
  url = {https://arxiv.org/abs/2606.28551},
}

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

Bib HTML PDF

@inproceedings{Hernandez2026Apertus,
  author = {Hernández-Cano, Alejandro and Hägele, Alexander and Huang, Allen Hao and Romanou, Angelika and Solergibert, Antoni-Joan and Pasztor, Barna and Messmer, Bettina and Garbaya, Dhia and Ďurech, Eduard Frank and Hakimi, Ido and Giraldo, Juan García and Ismayilzada, Mete and Foroutan, Negar and Moalla, Skander and Chen, Tiancheng and Sabolčec, Vinko and Xu, Yixuan and Aerni, Michael and AlKhamissi, Badr and Mariñas, Inés Altemir and Amani, Mohammad Hossein and Ansaripour, Matin and Badanin, Ilia and Benoit, Harold and Boros, Emanuela and Browning, Nicholas and Bösch, Fabian and Böther, Maximilian and Canova, Niklas and Challier, Camille and Charmillot, Clement and Coles, Jonathan and Deriu, Jan and Devos, Arnout and Drescher, Lukas and Dzenhaliou, Daniil and Ehrmann, Maud and Fan, Dongyang and Fan, Simin and Gao, Silin and Gila, Miguel and Grandury, María and Hashemi, Diba and Hoyle, Alexander and Jiang, Jiaming and Klein, Mark and Kucharavy, Andrei and Kucherenko, Anastasiia and Lübeck, Frederike and Machacek, Roman and Manitaras, Theofilos and Marfurt, Andreas and Matoba, Kyle and Matrenok, Simon and Mendonça, Henrique and Mohamed, Fawzi Roberto and Montariol, Syrielle and Mouchel, Luca and Najem-Meyer, Sven and Ni, Jingwei and Oliva, Gennaro and Pagliardini, Matteo and Palme, Elia and Panferov, Andrei and Paoletti, Léo and Passerini, Marco and Pavlov, Ivan and Poiroux, Auguste and Ponkshe, Kaustubh and Ranchin, Nathan and Rando, Javi and Sauser, Mathieu and Saydaliev, Jakhongir and Sayfiddinov, Muhammad Ali and Schneider, Marian and Schuppli, Stefano and Scialanga, Marco and Semenov, Andrei and Shridhar, Kumar and Singhal, Raghav and Sotnikova, Anna and Sternfeld, Alexander and Tarun, Ayush Kumar and Teiletche, Paul and Vamvas, Jannis and Yao, Xiaozhe and Zhao, Hao and Ilic, Alexander and Klimovic, Ana and Krause, Andreas and Gulcehre, Caglar and Rosenthal, David and Ash, Elliott and Tramèr, Florian and VandeVondele, Joost and Veraldi, Livio and Rajman, Martin and Schulthess, Thomas and Hoefler, Torsten and Bosselut, Antoine and Jaggi, Martin and Schlag, Imanol},
  title = {Apertus: Democratizing Open and Compliant LLMs for Global Language Environments},
  booktitle = {Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL)},
  url = {https://arxiv.org/abs/2509.14233},
  year = {2026},
}

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

Bib HTML PDF

@inproceedings{Bother2026Mixtera,
  author = {B{\"{o}}ther, Maximilian and Yao, Xiaozhe and Kerimoglu, Tolga and Graur, Dan and Gsteiger, Viktor and Klimovic, Ana},
  title = {Mixtera: A Data Plane for Foundation Model Training},
  booktitle = {Proceedings of the Conference on Management of Data (SIGMOD)},
  doi = {10.1145/3786668},
  url = {https://dl.acm.org/doi/10.1145/3786668},
  year = {2026},
}

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

Bib HTML PDF

@inproceedings{Bother2025Submod,
  author = {B{\"{o}}ther, Maximilian and Sebastian, Abraham and Awasthi, Pranjal and Klimovic, Ana and Ramalingam, Srikumar},
  title = {On Distributed Larger-Than-Memory Subset Selection With Pairwise Submodular Functions},
  booktitle = {Proceedings of the Conference on Machine Learning and Systems ({MLSys})},
  year = {2025},
  url = {https://arxiv.org/abs/2402.16442},
}

SIGMOD

Modyn: Data-Centric Machine Learning Pipeline Orchestration

Böther, Maximilian, Robroek, Ties, Gsteiger, Viktor and 4 more authors

In Proceedings of the Conference on Management of Data (SIGMOD) 2025

Bib HTML PDF

@inproceedings{Bother2025Modyn,
  author = {B\"{o}ther, Maximilian and Robroek, Ties and Gsteiger, Viktor and Holzinger, Robin and Ma, Xianzhe and T\"{o}z\"{u}n, P{\i}nar and Klimovic, Ana},
  title = {Modyn: Data-Centric Machine Learning Pipeline Orchestration},
  booktitle = {Proceedings of the Conference on Management of Data (SIGMOD)},
  year = {2025},
  doi = {10.1145/3709705},
  url = {https://dl.acm.org/doi/10.1145/3709705},
}