Dusan Bosnjakovic

Dusan Bosnjakovic is an AI Science group leader at Intuit working on agentic and security problems. At Data Con LA he presents independent research on detecting hidden, provider-level bias in large language models, the stable "lab signatures" that standard benchmarks miss. His method uses relative measurement and statistical techniques like variance decomposition to isolate how much of a model’s behavior traces to the organization that trained it rather than to the prompt.