Aarush Sah

I currently work at Meta Superintelligence Labs.

Previously, I led evals at Groq, where I started our evals efforts - building out our eval infrastructure, internal evals, customer evals, research, and more. While I was there, I also created openbench – an open-source standard for running evals easily, reliably, and in a reproducible manner. After Groq licensed its inference technology to NVIDIA, I continued my work there.

Before Groq, I was at Nous Research developing synthetic data pipelines for training language models.

When I'm not thinking about evals, I read epic fantasy novels – especially Brandon Sanderson's works – and optimize little parts of my life with software.