Dmytro Okhonko
Member Of Technical Staff at OpenAI
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolDmytro Okhonko is a seasoned machine learning and systems engineer with 10 years of experience building production-grade AI and low-latency infrastructure, now a Member of Technical Staff on OpenAI’s Sora team in Seattle. He helped scale speech recognition and NLP systems at Facebook AI, contributing notable open-source work to fairseq including a VGG-Transformer speech model and CTC loss integration. As a founding MTS at Samaya AI and former CTO at a high-frequency trading startup, he blends research-grade modeling with practical, high-throughput engineering. Early career roles at Microsoft and Samsung sharpened his GPU-accelerated graphics and image-processing expertise, while contributions to LogDevice reflect deep distributed-systems experience. Trained as a mathematician, he brings rigorous analytical thinking to messy production problems and a habit of shipping fixes that improve robustness under real-world constraints. Colleagues describe him as the kind of engineer who moves fluidly between research prototypes and hardened, scalable services.
10 years of coding experience
12 years of employment as a software developer
Master of Science (MS) Mathematics, Master of Science (MS) Mathematics at Taras Shevchenko National University of Kyiv
English, Ukrainian
Github Skills (11)
transformers10
pytorch10
machine-learning10
transformer10
speech-recognition10
artificial-intelligence10
nlp10
loss10
python10
fairseq10
ctc10
Programming languages (2)
Github contributions (5)
facebookresearch/fairseq
Mar 2019 - Oct 2019
Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
Role in this project:
ML Engineer
ML EngineerContributions:15 commits, 1 PR, 1 branch in 7 months
Contributions summary:Dmytro made several contributions related to the core functionality of the fairseq toolkit, particularly focusing on speech recognition models and loss functions. They added a new speech recognition model based on the VGG-Transformer architecture. Furthermore, the user incorporated Connectionist Temporal Classification (CTC) loss, a key component for training such models. Their work also included bug fixes related to input handling and batch size management.
pytorchnlpsequencepythontransformer-architecture
pytorch/audio
Jun 2019 - Jun 2019
Data manipulation and transformation for audio signal processing, powered by PyTorch
Contributions:5 comments in 6 days
pytorchaudio-processingmanipulationpythonsignal
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.