gal sapir
I’m a Staff Research Scientist at Pheno.AI, working on health AI. I design experiments, write code, and spend a lot of time on the question that comes before both: what would count as meaningful progress?
My background is in medicine and hyperpolarized magnetic resonance research. I’ve since worked as a software engineer and researcher, building systems and developing evaluations for health foundation models and agents.
At Pheno, I built PhenoBench, which turns questions about a deeply phenotyped human cohort into executable evaluations. I also helped define how we evaluate GluFormer, HealthFormer, and our health agent.
I’m interested in choosing the hills we ask AI systems to climb: which health problems are worth working on, what evidence would tell us we’re making progress, and how those questions might become environments where models can learn.
I write to work through these ideas at Sparse Thoughts.