Andrey Portnoy
Senior Software Engineer at NVIDIA
Canada
Summary
🤩
Rockstar🎓
Top SchoolAndrey Portnoy is a senior software engineer with a decade of experience building high-performance compilers, kernels, and tooling for modern ML hardware. He has driven performance and numerical-precision work on flagship open-source projects like JAX and TensorFlow/XLA, contributing GPU-focused optimizations, precision fixes for A100/H100, and tooling to visualize and dump computation graphs. At NVIDIA and Cerebras he has focused on DSLs, CUTLASS kernels, and wafer-scale compiler toolchains, repeatedly bridging low-level kernel work with higher-level runtime and profiling infrastructure. Comfortable across compiler internals, GPU performance engineering, and ML runtime instrumentation, he brings a mathematician’s rigor (BS Mathematics, UCSD) to production-grade systems and a track record of improving both correctness and measurable throughput.
10 years of coding experience
5 years of employment as a software developer
University of California, San Diego
Stackoverflow
1,479reputation
143kreached
47answers
5questions
Github Skills (28)
c-language10
python10
gpu-programming10
mlr10
tensorflow10
performance-optimization10
cuda10
xla10
jax10
cprogramming-language10
testing9
ml9
machine-learning9
deep-learning9
mle9
Programming languages (15)
Github contributions (5)
jax-ml/jax
Aug 2023 - Mar 2025
Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
Role in this project:
ML Engineer & Performance Engineer
ML Engineer & Performance EngineerContributions:53 reviews, 32 PRs, 82 comments in 1 year 6 months
Contributions summary:Andrey primarily contributed to the performance and stability of the JAX library, focusing on improving the numerical precision of computations. They addressed issues related to TF32 math on specific GPU architectures, such as A100 and H100, by modifying existing tests and introducing precision specifiers. Furthermore, the user optimized performance within the Mosaic GPU framework, adding tests, enhancing benchmarking scripts, and implementing CUPTI-based profiling for kernel analysis.
pytorchpythonjitautomatic-differentiationgpu
tensorflow/tensorflow
Apr 2019 - Sep 2019
An Open Source Machine Learning Framework for Everyone
Role in this project:Back-end Developer
Back-end DeveloperContributions:5 comments in 5 months
Contributions summary:Andrey primarily contributed to extending the XLA Python interface, enabling the dumping of computation subgraphs to HTML. This involved modifying the Python interface to include methods for rendering computations. Additionally, the user worked on improving the HLO dumper by integrating the Google CDN for external resources, and instrumenting code to dump IR after passes in the LowerToXlaGpuRuntime stage. Furthermore, the user made modifications to support unsigned MLIR integers during FFI attribute conversion and overloaded the operator<< for XLA_FFI_DataType.
pythondata-sciencedeep-learningmlmachine-learning
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.