Picture of me

Oliver Rausch

I build stuff related to compilers, deep learning, high performance computing and systems.

I’m currently doing systems research at Anthropic in San Francisco. We work on scaling deep learning for AI safety and alignment research.

I previously studied computer science, at ETH Zurich (2017—2021) and Oxford (2021—2022). Contact me on github, email or linkedin.

Prior Projects

Scalable and performant deep learning (2020—2022): With the Scalable Parallel Computing Laboratory, I worked on daceml, a really fast ML framework/compiler. Its visual, interactive workflow makes it the first framework that “puts the performance engineer in the driver’s seat”.

We competed with the major deep learning frameworks (PyTorch, TF, JAX, etc.) and attained the fastest performance for a range of models.
The growing documentation burden in healthcare (2020—2022): I was a co-founder at cognote.ai, where we worked on automating clinical documentation using speech recognition + deep NLP. We assembled and labeled a large german dataset for this domain in collaboration with the LMU Klinkum.

Microsoft logo

Microsoft Research (Cambridge, UK): worked on scalable neural rendering and architectural extensions to ONNX Runtime with Tim Harris and the FastNeRF authors.

ETHZ logo

Scalable Parallel Computing Laboratory (Zurich, CH): worked towards a paper submission for daceml with Tal Ben-Nun and Torsten Hoefler.

Oracle Labs (Zurich, CH): developed a web-scale, graph neural network anomaly detection pipeline and PGX.ML, a graph machine learning library, with Rhicheek Patra and Damien Hilloulin.

Araneum logo

Araneum Technologies (Zurich, CH): worked on applied graph machine learning in the banking sector.