Jure Zbontar

Ljubljana, Slovenia

Join Prog.AI to see contacts

Summary

🤩

Rockstar

🎓

Top School

Jure Zbontar is a seasoned software engineer and research-focused ML practitioner with 21 years of experience, currently working as a Research Engineer at OpenAI after roles at Meta and Tesorai. He holds a PhD in Computer Science from the University of Ljubljana and combines deep academic training with hands-on engineering—shipping core training and evaluation code, mixed-precision and distributed training fixes, and CUDA kernels for high-performance ML. His open-source contributions include work on prominent projects such as Facebook Research’s Barlow Twins and the mc-cnn stereo-matching repo, plus model and algorithm integrations for the Orange data analysis ecosystem. Jure excels at bridging research prototypes to production-ready implementations, with a practical focus on optimization, data pipelines, and reproducible evaluation. Based in Ljubljana, he brings a pragmatic yet research-driven approach to complex ML systems and deployment challenges. Less obvious: he has experience across both Python and lower-level CUDA/Lua toolchains, enabling end-to-end performance improvements.

21 years of coding experience

5 years of employment as a software developer

Doctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at University of Ljubljana, Faculty of Computer and Information Science

Github Skills (28)

pytorch10

distributed-training10

python10

data-mining10

machine-learning10

mlp10

mask-rcnn10

regression10

cuda10

computer-vision10

faster-rcnn10

documentation10

image-processing9

evaluation9

numpy9

Programming languages (5)

C++LuaJupyter NotebookPythonCuda

Github contributions (5)

jzbontar/mc-cnn

Sep 2015 - Aug 2017

Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches

Role in this project:

ML Engineer

Contributions:37 commits, 24 pushes, 1 branch in 1 year 11 months

Contributions summary:Jure primarily contributed to the core functionality of the `mc-cnn` repository, which focuses on stereo matching using convolutional neural networks. Their commits include modifications to the main training and prediction scripts (`main.lua`, `predict_kitti.lua`), data loading, and preprocessing, as well as implementing supporting CUDA kernels (`adcensus.cu`) and scripts for data handling (`download_middlebury.sh`). They also addressed bugs and made improvements to the overall system, demonstrating a focus on the implementation and performance of the stereo matching models.

convolutional-neural-networks

facebookresearch/barlowtwins

Mar 2021 - Dec 2021

PyTorch implementation of Barlow Twins.

Role in this project:

ML Engineer

Contributions:10 commits, 8 pushes, 34 comments in 9 months

Contributions summary:Jure primarily contributed to the core training and evaluation components of the Barlow Twins implementation. They made changes to the main training loop, including adjusting learning rates and integrating mixed-precision training with `torch.cuda.amp`. Further modifications involved fixing issues related to PyTorch Hub integration and distributed data parallel wrapping in the evaluation script, demonstrating an understanding of model deployment and optimization strategies. Moreover, the user updated the model's pre-trained weights and parameters from available external sources.

pytorch

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.

Request Free Trial