Oana works primarily on explainable neural networks, building models that generate human-like explanations for their predictions.