Igor Sadalski

I’m a Machine Learning Scientist at Cellular Intelligence, a Harvard-MIT spin-off building multi-modal foundation models for life sciences ($60M+ raised, led by Khosla Ventures). As the 6th engineer, I design and implement pre-training and post-training pipelines for billion-parameter transformers on terabytes of data across 32 AMD MI325x GPUs using PyTorch FSDP and SLURM. I’ve co-authored publications at ICML, ICLR, and Nature Machine Intelligence on pre-training strategies and measurement noise scaling laws for foundation models.

I hold a Master of Science with Distinction in Computer Science (Machine Learning) from Imperial College London. I’ve previously conducted research at Harvard University (with Prof. Stephanie Gil) on transformer attention mechanisms for spatio-temporal forecasting, and at Caltech (with Prof. Aaron Ames) on CVAEs and diffusion models for safe robotic control.

I was selected as a Fulbright Scholar from Poland (one of only three recipients) — which I declined to work in industry. In 2020, I was the top student in Poland, achieving a perfect score in matriculation exams.

In free time I like to cook and play board games with my friends.