LLMs have rapidly become the driving force behind AI applications across various fields, including healthcare. Their use in digital health ranges from medical question-answering to more advanced systems that assist with diagnostics and clinical decision-making.

Our research in this area focuses on analyzing LLM-generated outputs from both linguistic and factual accuracy perspectives, ensuring that responses are coherent, precise, and contextually appropriate. We develop modular Retrieval-Augmented Generation (RAG) pipelines that enhance patient-physician communication by grounding responses in verified medical knowledge. Additionally, we design multi-agent systems, including AI-driven prototypes for delivering digital therapies to support mental health 

The ability to integrate and reason across diverse sensory inputs is fundamental to human and artificial intelligence. Multi-Modal Knowledge Graphs (MMKGs) extend traditional Knowledge Graphs by associating entities with different modal representations, such as text, images, audio, and videos, to better capture their semantics.  

Our research in this area focuses on designing a novel ontology design pattern that formalizes the relationship between an entity (and the information it conveys), whose semantics can have different manifestations across different media, and its realization in terms of a physical information entity. 

We aim to develop a unified approach to MMKGs, which requires a formalized structure that harmonizes different multi-modal ontologies, ensuring cross-field integration and broader applicability.

Hybrid AI models, particularly neuro-symbolic approaches, combine symbolic reasoning with sub-symbolic techniques to leverage the strengths of both paradigms. These models enhance performance and interpretability, making them valuable for complex AI applications. 

Our neuro-symbolic research focuses on two main directions: improving the reasoning capabilities of fuzzy neuro-symbolic systems and developing a scalable hybrid framework that integrates symbolic fuzzy inference systems with neuro-symbolic fuzzy logical frameworks and evolutionary algorithms.

The main goal is to enable efficient processing of large-scale real-world data while maintaining accuracy, transparency, and interpretability.

The rapid evolution of new technologies—Artificial Intelligence included—demands proactive ethical reflection integrated from the earliest stages of development, especially in healthcare. The “ethics by design” approach ensures that innovations align with fundamental values by embedding ethical considerations into technological design. AI in healthcare domain raises both new and longstanding ethical challenges: a multidisciplinary, collaborative dialogue is essential to addressing these issues holistically while keeping human well-being at the core of healthcare innovation.

Our research interests in the ethics area include the assessment of bias in LLMs, the analysis of medical conversations to mitigate the risk of misunderstanding and ensuring fairness and algorithmic transparency in multi-agent conversational systems.