About me

I am a Ph.D. student in Data Science under the supervision of Prof. Fabrizio Silvestri, and also in collaboration with the Institute of Information Science and Technologies "Alessandro Faedo", the largest CNR Institute for Computer Science. 

I am pursuing my Ph.D. at DIAG (Dipartimento di Ingegneria Informatica, Automatica e Gestionale).

My research started with Graph Neural Networks (GNNs): how to design them, understand their behavior, and improve their generalization across unseen graphs. Over time, I began to feel that graph learning progress was reaching a plateau. This motivated me to explore the potential of existing GNNs across different applications, particularly real-world problems where graph structure can make a meaningful difference.

This direction led me to Graph Retrieval-Augmented Generation (RAG) and also the alignment of graph encoders with language model decoders. I am interested in how we can compress the information contained in a textual graph into representations that a powerful text decoder can interpret and use. I believe taht this connection remains poorly understood and presents challenging questions about what information is preserved, how it is represented, and how a decoder makes sense of it.

Working on this connection has broadened my interest in alignment across data modalities and led me to a more general question: how can we connect external knowledge, in any modality, to a generative model whose outputs may belong to another modality? Beyond learning compatible representations, a central challenge is deciding which external information is relevant, especially when relevance is not explicitly defined or supervised.

I have explored this question in cellular perturbation modeling, using differentiable RAG to learn relevance implicitly through the prediction task. As external knowledge grows in scale and diversity, I see learned retrieval as a crucial bridge between the information available and the context a generative model needs. My broader goal is to understand how models can learn both what knowledge matters and how to use it across modalities.

I hold an M.Sc. in Artificial Intelligence and Robotics (LM-32) and a B.Sc. in Management Engineering (L-8), both from Sapienza University of Rome.

Curriculum Vitae

• Updated September 2026