Roki Seydi
Clinical AI Safety · Evaluation Infrastructure · Care Delivery Intelligence
I build evaluation infrastructure and adversarial simulation for clinical AI safety — testing whether AI systems produce reasoning that holds under the conditions that matter most: real patients, constrained systems, and contexts that benchmarks don't reach.
Alongside open-source evaluations, I'm building Elost: a care pathway simulation engine for adversarial testing of clinical AI against the patients most likely to be dismissed.