commit 2f8654b (2025-08-03 20:42:43 -0400) Torsten Scholak: revamp
I'm Torsten Scholak. I lead the Foundation Model Lab at ServiceNow Research, where I specialize in turning GPU compute into shipped language models. I co-founded the ServiceNow Language Model (SLAM) initiative and currently focus on the Apriel model family, which are open-weight LLMs built for enterprise agentic tasks and efficient inference. Apriel-5B and Apriel-Nemotron-15B-Thinker set state-of-the-art marks on enterprise benchmarks while cutting serving costs and latency. This work was highlighted by Jensen Huang during the ServiceNow Knowledge 2025 keynote.
Previously, I pioneered grammar-constrained decoding with the PICARD text-to-SQL parser (400+ citations, ex-Spider SOTA), co-organized Deep Learning for Code at ICLR 2022/2023, and helped build Hasktorch to bring GPU-accelerated machine learning to Haskell.
Before all that: I trained as a theoretical physicist in Germany (Bayreuth and Freiburg, Ph.D. 2011), then moved to Canada for a postdoc at the University of Toronto, before turning to industry at Unata and Element AI. Element AI was acquired by ServiceNow in 2021.
If you want the full paper trail: resume, publications.
That's it. If you actually made it to the end, check the links up top and maybe say hi: contact.