Marcin Junczys-dowmunt
Principal Research Scientist Manager at Microsoft
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolMarcin Junczys‑dowmunt is a Principal Research Scientist and manager at Microsoft leading research for Microsoft Translator with 14 years of applied machine translation and NLP experience. He combines academic rigor—from a PhD and faculty role at Adam Mickiewicz University—to large-scale production ML, having consulted for WIPO, the UN, Samsung and contributed to fast, efficient open-source toolchains. Marcin is the driving force behind the Marian NMT project and has deep systems-level expertise (C++/TensorFlow) improving batching, pruning and compact data structures in widely used projects like KenLM, Moses and Nematus. His work bridges neural methods, automatic post-editing and grammatical error correction while prioritizing performance and deployability in real-world, high-throughput settings. Based in Seattle, he is equally comfortable shipping production optimizations and publishing research, a rare blend that accelerates translation quality at scale.
14 years of coding experience
11 years of employment as a software developer
MA German Philology, MA German Philology at Uniwersytet Kazimierza WielkiegoPh.D Applied Linguistics, Ph.D Applied Linguistics at Uniwersytet im. Adama Mickiewicza w PoznaniuBSc Computer Science, BSc Computer Science at College of Computer Science in Łódź
Polish, German, English
Github Skills (25)
algorithm10
algorithms10
machine-translation10
c-language10
neural-machine-translation10
python10
multithreading10
data-structure10
pruning10
tensorflow10
language-modeling10
data-structures10
cprogramming-language10
performance-analytics9
performance-monitor9
Programming languages (7)
Github contributions (5)
marian-nmt/marian
Apr 2016 - Feb 2022
Fast Neural Machine Translation in C++
Role in this project:
Back-end Developer
Back-end DeveloperContributions:11 releases, 2356 commits, 11 PRs in 5 years 10 months
Contributions summary:Marcin primarily focused on improving the efficiency of the data processing in the `marian-nmt/marian` repository by enhancing the batch generation process. They implemented features like batch read-ahead using multi-threading. They also worked on incorporating a set of tools to handle the compilation errors to make it run more reliably in Windows.
machine-translationtranslationneural-machine-translationcudagpu
moses-smt/mosesdecoder
Aug 2012 - Mar 2017
Moses, the machine translation system
Role in this project:Backend Developer
Backend DeveloperContributions:432 commits, 86 pushes, 4 branches in 4 years 8 months
Contributions summary:Marcin contributed to the core functionality of the Moses machine translation system by adding source code for a compact phrase table. The commits detail the implementation of a `PhraseTableCreator` class in C++, which involves creating source phrase indices, encoding target phrases, and compressing them using Huffman coding. The work involved modifying several `.cpp` and `.h` files, showcasing expertise in data structures, encoding techniques, and efficient algorithm design for machine translation tasks.
i18nmachine-translationtranslationlocalizationmoses
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.