Emanuele Fontana is a Ph.D. Student at the Computational Intelligence Laboratory (CILab), University of Bari Aldo Moro. His research focuses on trustworthy foundation models and mechanistic interpretability, with particular emphasis on the real-time detection and mitigation of hallucinations in Large Language Models.
His work investigates latent-space alignment and shared representations across different model architectures, with the goal of identifying transferable truthfulness features and developing lightweight detection and steering mechanisms that do not require model-specific retraining, external retrieval systems, or multiple inference passes.
His research also extends these ideas to Vision-Language Models, investigating multimodal hallucinations and the alignment between visual and textual representations toward efficient and reusable safety mechanisms for foundation models.