i'm serving as vp of research at autopoiesis sciences, where we're training models that can do science autonomously. our mission is to be the first to create ai systems capable of out-of-domain breakthrough discoveries.

my research focuses on how metacognitive capabilities are configured and can be steered in latent spaces of open frontier models. i believe that scaling rl will not lead to models capable of original thinking and breakthrough discoveries. rl rewards convergence and conservatism, which are the opposite of discovery. scientists make decisions under uncertainty by reasoning about sparse evidence from first principles instead of relying on a verifier. i'm working on a new paradigm to distill that judgment into ai models.

i'm a humboldt university of berlin valedictorian who also holds a master's degree in data science from uc berkeley. during my master's, i led a team in winning the google case competition by building a global staffing optimizer using time series modeling, queuing theory, monte carlo simulation, and optimization algorithms.

contact: linkedin / x / email