Brian Coutinho
Principal Software Engineer at Facebook
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolBrian Coutinho is a Principal Software Engineer with eight years of experience building high-performance systems and tooling for AI at scale, currently focused on accelerating GPU workloads at NVIDIA. He previously led system observability and profiler efforts at Meta, co-maintained PyTorch Profiler, Kineto and Dynolog, and built fleet-wide GPU trace analysis and datasets that inform capacity planning and kernel optimization. With roots in CPU microarchitecture from Apple and verification experience at ARM, he brings rare cross-stack expertise from silicon validation to datacenter monitoring. An active open-source contributor, his PyTorch work added richer CPU/GPU correlation and tracing features that improve real-world performance analysis for deep learning. He combines hands-on engineering with a systems-level view that turns low-level instrumentation into actionable insights for AI infrastructure.
8 years of coding experience
4 years of employment as a software developer
Master’s Degree Electrical Engineering, Master’s Degree Electrical Engineering at University of Wisconsin-MadisonBachelor of Technology (BTech) Electrical Engineering, Bachelor of Technology (BTech) Electrical Engineering at Indian Institute of Technology Jodhpur
Github Skills (15)
cuda10
pytorch10
machine-learning10
deep-learning10
gpu10
profiler10
python10
performance-monitoring9
performance-measurement9
autograd9
performance-analytics9
performance-monitor9
performance-analysis9
performance-tuning9
tensor8
Programming languages (3)
Github contributions (5)
pytorch/pytorch
Nov 2022 - Jan 2023
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
ML Engineer
ML EngineerContributions:57 reviews, 5 commits, 60 PRs in 2 months
Contributions summary:Brian's commits primarily focused on enhancing the PyTorch profiler, a tool for analyzing and optimizing deep learning models. They implemented features to track and manage step counts from multiple sources within the profiling process, improving the accuracy of performance measurements. The user also added functionality to enable the inclusion of CPU activity alongside CUDA profiling, enabling users to correlate CPU and GPU operations. Furthermore, they introduced the option to include CUDA synchronization events in the trace and added new metrics for the PyTorch profiler to measure time and other statistics, enabling better performance analysis.
pythongpu-accelerationdeep-learninggpunumpy
briancoutinho/pytorch
Feb 2022 - Nov 2024
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Contributions:49 pushes, 12 branches in 2 years 9 months
pythongpu-accelerationdeep-learninggpuacceleration
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.