Dmytro Pykhtar
Senior Machine Learning Engineer at NVIDIA
Kyiv, Ukraine
Summary
🤩
Rockstar🎓
Top SchoolDmytro Pykhtar is a Senior Machine Learning Engineer based in Kyiv with seven years of experience building and productionizing large-scale ML models at NVIDIA. He has progressed from research internships to senior engineering roles, contributing to flagship projects like NVIDIA NeMo where he enhanced model configuration and parallelism support for BERT and added unit tests to harden deployment pipelines. With a strong statistical foundation from a master’s degree in Statistics, he bridges theory and engineering—tuning hyperparameters, calculating model sizes, and shaping training configurations for efficient distributed training. Colleagues know him for pragmatic contributions that improve both developer ergonomics and runtime efficiency, and for paying close attention to validation and reproducibility. He combines deep technical fluency in model-parallel training with hands-on experience shipping cloud-native and on-prem ML pipelines.
7 years of coding experience
4 years of employment as a software developer
Master's degree, Statistics, Master's degree, Statistics at Taras Shevchenko National University of Kyiv
English, Ukrainian
Github Skills (11)
hyperparameter-optimization10
machine-learning10
custom-configuration10
configurations10
deep-learning10
yml-configuration10
system-configuration10
python10
bert10
pytorch8
tensorflow8
Programming languages (3)
Github contributions (5)
NVIDIA/NeMo-Framework-Launcher
Nov 2022 - Mar 2023
Provides end-to-end model development pipelines for LLMs and Multimodal models that can be launched on-prem or cloud-native.
Role in this project:
ML Engineer
ML EngineerContributions:3 releases, 103 reviews, 17 commits in 4 months
Contributions summary:Dmytro focused on enhancing the model configuration tools for the Nemo framework, specifically for BERT models. They implemented support for varying BERT model sizes, including configurations for global batch size, tensor parallelism, and pipeline parallelism. The user also made adjustments to model size calculations and training configurations, including updates related to the calculation of model parameters and the selection of hyperparameters like learning rates. Further, they also wrote unit tests to validate the configurations.
Contributions:3 pushes, 1 branch in 1 year 8 months
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.