Ezinwanne (Ezi) Ozoani is Co-Founder of Aethon, where she leads the design and deployment of reinforcement learning–driven financial intelligence systems, spanning research, infrastructure, and live production environments. Previously, she led ML ethics and red-teaming efforts at Hugging Face, contributed to multimodal and large-scale model evaluation, and collaborated with institutions including MIT and Harvard on applied AI in healthcare and low-resource settings. Her research spans reinforcement learning, interpretability, and safety, with a focus on translating theoretical advances into production-grade systems. At Aethon, her team has explored constrained RL in high-stakes environments such as financial decision-making, uncovering failure modes that only emerge post-deployment. This work informs her perspective on controlling optimization directly—moving beyond reward shaping toward structural guarantees. Her work sits at the intersection of RL, systems design, and AI safety, with an emphasis on building models that behave reliably under pressure. Additionally, she works with the EU AI Office on the Codes of Practice, operating at the intersection of regulation and applied research.
