Andrea Santilli is a Research Scientist at Nous Research and holds a PhD in Computer Science from GLADIA at Sapienza University of Rome. His research focuses on large language models (LLMs), robustness, reliability, and multimodal learning. He previously worked at Apple MLR, Hugging Face’s BigScience, and Pi School. He earned his MSc and BSc in Computer Science from Tor Vergata University and Sapienza. Education: PhD in Computer Science, Sapienza University of Rome (2024) MSc in Computer Science, University of Roma Tor Vergata (2020) BSc in Computer Science, University of Roma Tor Vergata (2018) Research Interests: Santilli’s work spans LLM robustness , mechanistic interpretability , multimodal neural databases , and instruction-tuning . He introduced Parallel Jacobi Decoding and contributed to projects like BLOOM, Camoscio, and Fauno. His research bridges syntax-aware NLP, privacy-preserving LLMs, and cross-modal alignment. Publications: His work includes advancements in 3D-text latent space alignment (CVPR 2025), evolutionary merging (ICML 2025), and efficient decoding (ACL 2023). Over 15+ peer-reviewed papers span venues like ACL, CVPR, and ICLR. Awards: Received the Emanuele Pianta Award for his MSc thesis on continual language learning with syntax-based episodic memory. Grants & Projects: Winner of ‘Machine Learning Algorithms for Translation’ grant (2022), developing Parallel Decoding Co-PI for ‘Multimodal AI for 3D Analysis’ (2021) with Ecole Polytechnique Labs & Teams: Active in GLADIA (Sapienza), Apple MLR, and Hugging Face’s BigScience initiative. Core contributor to open-source projects like PromptSource and BLOOM.








