Matteo Bonotto, Ph.D.

I am a Senior AI Research Engineer specialized in large language models, high-performance deep learning, and scientific computing. My work spans the LLM stack, from model architecture and training to GPU optimization, distributed systems, evaluation, large-scale inference and agentic harness design.

At Giotto.ai, I work on core LLM research and engineering. My current work includes designing and implementing custom Transformer architectures, distributed training on multi-node GPU clusters, GPU kernel optimization with Triton, and scalable LLM inference across NVIDIA and AMD accelerators. I also work on reasoning and agentic systems, model evaluation, quantization, and long-context inference. I contributed to solutions awarded gold medals in the Kaggle ARC Prize in 2024 and 2025.

Before moving full-time into AI research, I worked in magnetic-confinement fusion, the field in which I obtained my Ph.D. (my thesis). My research focused on mathematical and numerical modeling of plasma physics, electromagnetics, plasma equilibrium, MHD, and plasma control. I’ve been working on mathematical modeling for electromagnetics, plasma equilibrium and magnetohydrodynamics (here some papers). I’ve been developing surrogate models of plasma equilibrium using physics-informed neural operators (see PlaNet and this dataset), and I developed a FEM-based tool for plasma equilibrium with a novel formulation for boundary conditions (FRIDA), and many additional things!