Jacek Czaja
Machine Learning Engineer at Intel Corporation
Gdańsk, Pomeranian Voivodeship, Poland
Summary
🤩
Rockstar🎓
Top School
Top expert inDeep Learning and Computer Vision Technologies
Jacek Czaja is a Machine Learning Engineer with over a decade of experience building and optimizing deep learning frameworks and Linux-based systems, currently applying his skills at Intel in Gdańsk. He specializes in performance engineering—particularly MKL-DNN/oneDNN optimizations and bfloat16 support—combining C++, Python, and Rust to squeeze latency and throughput gains on Intel architectures. His background spans graphics driver and OpenGL/OpenCL development through to production ML workloads, giving him rare expertise at the intersection of low-level systems and ML stacks. An active open-source contributor, he has documented and implemented critical optimizations in the widely used PaddlePaddle project, and maintains both hobby and work-related GitHub repos. Not actively seeking new roles, he nevertheless explores Rust and ML as ongoing personal interests, reflecting a continuous drive to modernize high-performance code.
10 years of coding experience
6 years of employment as a software developer
Master Engineer, Computer Science, Master Engineer, Computer Science at Gdańsk University of TechnologyI LO im. ks. Elzbiety
Github Skills (9)
paddlepaddle10
c-language10
deep-learning10
edn10
cprogramming-language10
performance-optimization10
documentation10
technical-writing10
tensorflow4
Programming languages (7)
Github contributions (5)
PaddlePaddle/Paddle
Feb 2018 - Dec 2022
PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice （『飞桨』核心框架，深度学习&机器学习高性能单机、分布式训练和跨平台部署）
Role in this project:
ML Engineer
ML EngineerContributions:661 reviews, 174 commits, 416 PRs in 4 years 10 months
Contributions summary:Jacek's primary contributions focused on enhancing the performance of the PaddlePaddle deep learning framework. The commits demonstrate the implementation of specific optimizations, particularly within the MKLDNN (Intel Math Kernel Library for Deep Neural Networks) backend, including integration for operations like Softmax and the addition of support for the use of bfloat16 data type. Their work included adding and improving the usage of the MKLDNN-optimized implementations of operators like Elementwise, Convolution, and Batch Norm, which contribute to the framework's efficiency on Intel architectures, and added new features like support for residual connections. The commits also included addressing bugs, and code style fixes, which improved the framework's functionality and performance.
pytorchpythonparalleldeep-learningpaddlepaddle
PaddlePaddle/docs
Sep 2019 - May 2022
Documentations for PaddlePaddle
Role in this project:Technical Writer
Technical WriterContributions:1 review, 7 commits, 6 PRs in 2 years 8 months
Contributions summary:Jacek's contributions primarily focus on creating documentation for the PaddlePaddle project, specifically regarding MKL-DNN and oneDNN integration. They have written design documents detailing topics such as data transformation, acquire API, object caching, NHWC layout support, and in-place execution. Furthermore, the user documented the oneDNN GRU operator and the keys used for the oneDNN cache. The changes consist of adding and modifying `*.rst` files, demonstrating a focus on technical documentation.
deep-learningpaddlepaddledocumentationspaddlepaddle-tutorials
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.