I am a Research Scientist in Natural Language Processing and Large Language Models. I hold a PhD in Computer Science from GLADIA at Sapienza University of Rome, where my doctoral research focused on building effective, efficient, and reliable Large Language Models. Previously, I worked as a Research Scientist at Nous Research and at Apple in the MLR team (see Experiences for the full list).

My current research interests center on improving language models’ robustness and reliability through uncertainty estimation and mechanistic interpretability (see pub1 and pub2). In the past, I have worked on a wide range of topics, including syntax in transformers (see KERMIT), efficient decoding methods (we introduced Parallel Jacobi Decoding to roughly double decoding speed, now adopted by lmsys), and instruction tuning for LLMs (we introduced instruction tuning, which is now a standard component of modern LLM training pipelines). I have also worked on instruction tuning for the Italian language (see Camoscio), as well as related areas such as preserving privacy in LLMs, audio LLMs, and multimodal neural databases. A full list of publications is available on my Google Scholar profile.

In the news: The BigScience project, to which I contributed, has been covered by outlets such as MIT Technology Review and The Washington Post. I was also featured in La Repubblica as one of the “500 Italians who matter in AI” (article in Italian). More recently, our work on LLM injectivity (aka the Pringle paper) received broad attention with roughly 5 million views!

If you would like to connect, feel free to reach out on X, LinkedIn, or through the contact form below.