MrCosmo Santoni

PhD candidate in Foundational Machine Learning and Applied Mathematics at Imperial College London, working at the intersection of machine learning research, research software engineering, and production ML systems. My work focuses on state-space architectures, neural emulation, and simulation-based inference—achieving 125,000× speedups over traditional simulators while maintaining calibrated uncertainty estimates.

My research is deployed in production, specifically the neural emulators I developed support World Health Organisation malaria intervention planning across Africa, and my earlier work on real-time epidemic inference directly informed UK government COVID-19 policy, for which I received the SAGE Award from the UK Chief Scientific and Medical Officers.

Publications include Nature and The Lancet, with recent work presented at NeurIPS 2025 and manuscripts targeting ICML 2026 and KDD 2026.

Beyond academic research, I contribute to open-source ML infrastructure. My JAX/Flax implementation of Mamba-2 was merged into Google's official jax-ml/bonsai repository through an ongoing collaboration with the Google JAX team. I also work on LLM robustness and safety, including metacognitive studies shared with Anthropic Trust & Safety.

My research is primarily applied to Infectious Diseases and time-series.