Andrzej Kotłowski
Deep Learning Software Engineer at Intel Corporation
Gdynia, Pomeranian Voivodeship, Poland
Summary
🤩
Rockstar🎓
Top SchoolAndrzej Kotłowski is a Deep Learning Software Engineer with 10 years of experience applying performance engineering to high‑performance ML frameworks and graphics systems. Based in Gdynia, Poland, he has spent over a decade at Intel focusing on backend optimization, stability fixes and operator fusions that make deep learning workloads run faster and more reliably in production. His notable open-source contributions include performance and synchronization fixes for the widely used MXNet framework and multithreaded COO-to-CSR optimizations for the DGL graph library, demonstrating a practical knack for squeezing performance from complex data paths. He blends low-level systems understanding with ML operator knowledge, enabling impactful integrations like oneDNN for quantized operator speedups. Colleagues rely on him to diagnose subtle concurrency and compilation issues that often hide in large tensor and multi-threaded code paths.
10 years of coding experience
9 years of employment as a software developer
Gdańsk University of Technology
Github Skills (16)
sparse-matrix10
sparse-array10
c-language10
deep-learning10
cprogramming-language10
edn10
sparse-file10
performance-optimization10
graph-neural-network10
python10
multithreading10
unit-test7
neural-network5
convolutional-neural-networks5
pytorch5
Programming languages (4)
Github contributions (5)
apache/mxnet
Apr 2020 - Sep 2022
Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
Role in this project:
Back-end Developer & Performance Engineer
Back-end Developer & Performance EngineerContributions:182 reviews, 29 commits, 32 PRs in 2 years 5 months
Contributions summary:Andrzej primarily contributed to improving the performance and stability of the `mxnet` deep learning framework. Their work included fixing synchronization issues in the BatchNorm layer and addressing compilation problems with large tensors when using MKL. They also implemented microbenchmarks for FC operations and integrated the oneDNN library to optimize operators, particularly focusing on fusing and optimizing the quantized operators to enhance performance.
pythonschedulerdataflowmutationdata-science
dmlc/dgl
Jan 2023 - Jan 2023
Python package built to ease deep learning on graph, on top of existing DL frameworks.
Role in this project:Back-end Developer / Performance Engineer
Back-end Developer / Performance EngineerContributions:35 reviews, 1 commit, 7 PRs in 1 day
Contributions summary:Andrzej's primary contribution focused on optimizing the performance of the `dgl` library, specifically improving the COO to CSR conversion process. This involved modifying the implementation to leverage multi-threading for improved performance and address performance bottlenecks in the COOToCSR implementation. The user also made several adjustments relating to threads and data processing. The user also made the testing reproducible with seeding.
pytorchpythondeep-learningmachine-learninggraph-neural-networks
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.