Mario Casado
Member Of Technical Staff at OpenAI
United Kingdom
Summary
🤩
Rockstar🎓
Top School
Top expert inArtificial Intelligence and Machine Learning Technologies
Mario Casado is a Principal Software Engineer and compiler-savvy ML systems lead with 11 years of experience, currently a Member of Technical Staff at OpenAI after leading the PyTorch core dev team at Quansight. He specializes in high-performance back-end engineering for deep learning runtimes, contributing performance-critical work to PyTorch (torch.linalg, torch.compile, autodiff) and to the Triton compiler—evidenced by optimizations like vectorised norms, memory-efficient linalg.solve, and compiler codegen improvements. With a DPhil in Computational and Applied Mathematics from Oxford, he blends formal academic training with production-grade systems design, often operating where compilers and ML meet (“compilers go brrrr”). He advises startups on applying AI to novel domains such as blockchain data, and is comfortable driving cross-organisational collaboration with Meta and other ecosystem partners.
11 years of coding experience
4 years of employment as a software developer
DPhil (PhD), Computational and Applied Mathematics, DPhil (PhD), Computational and Applied Mathematics at University of OxfordBachelor's degree, Mathematics, Bachelor's degree, Mathematics at Universidad Complutense de Madrid
Github Skills (13)
cuda10
pytorch10
automatic-differentiation10
triton10
python10
back-end-development10
linear-algebra10
compiler-development10
code-optimization9
c-language8
cprogramming-language8
machine-learning7
deep-learning7
Programming languages (14)
Github contributions (5)
pytorch/pytorch
Mar 2021 - Jan 2023
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:4324 reviews, 1984 commits, 483 PRs in 1 year 11 months
Contributions summary:Mario made several contributions to the PyTorch library, focusing on improving the functionality and efficiency of core linear algebra operations. Their work included making improvements to the decomposition of `linalg.solve` and implementing `linalg.vecdot`. They also made a number of performance improvements, such as vectorising `norm` and optimising memory usage in `linalg.solve`. In addition, they worked on the automatic differentiation implementation for a number of functions.
gpu-accelerationneural-networkpythonautogradgpu
triton-lang/triton
Mar 2023 - Jul 2026
Development repository for the Triton language and compiler
Role in this project:Back-end Developer
Back-end DeveloperContributions:890 reviews, 321 PRs, 737 pushes in 3 years 5 months
Contributions summary:Mario contributed to the Triton language and compiler, specifically focusing on the back-end code. Their commits demonstrate an understanding of compiler internals, including improvements to the code generation and optimizations within the TritonGPU compiler. The contributions include implementing features like supporting scalar conditions in device assertions, removing unused parameters, and improving error messages related to fp8 types. These changes focused on improving compiler functionality and code efficiency.
compiler
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.