i'm emilia,
and i do research on
multi-agent llms.

hi, i'm emilia. i'm a phd student in the ellis program, split between ellis alicante and the university of toronto, advised by nuria oliver and zhijing jin.

my thesis is about verifying multi-agent work: when a pile of agents hands you a result, how do you actually know they did the job right? humans get tired and miss things, model monitors have their own blind spots, and hybrid protocols sit somewhere in between. i build benchmarks that measure the verifier rather than the agent.

before the phd i worked as an ml engineer and researcher: llm safety and polish-language models at nask (including pllum, poland's first open-source llm), security ml at google, on-device nlp at samsung, and clinical llm systems at jutro medical.