Vladimir Karpukhin
Staff Software Engineer at Google
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolVladimir Karpukhin is a Staff Software Engineer in Seattle with seven years of focused experience delivering production-grade ML and data systems across major tech and research organizations. He has progressed from search engineering and content enrichment roles to senior ML leadership at Meta, Forethought, and AI-focused startups, and now contributes at Google, blending research rigor with product-minded execution. His open-source contributions to Facebook Research’s Dense Passage Retriever highlight practical expertise in retrieval systems and data pipelines, including multi-task training and representation selection for biencoders. Comfortable leading cross-functional teams, he specializes in scaling AI platforms and rebuilding data flows for robust training and inference. Trained as a master in applied mathematics and physics at MIPT, he brings a strong analytical foundation and a knack for turning complex ML research into reliable engineering.
7 years of coding experience
17 years of employment as a software developer
master applied mathematics and physics, master applied mathematics and physics at Moscow Institute of Physics and Technology (State University) (MIPT)
Github Skills (8)
python10
data-engineering10
data-pipelines9
data-pipeline9
machine-learning9
nlp8
pytorch8
natural-language-processing8
Programming languages (1)
Github contributions (4)
facebookresearch/DPR
May 2020 - Mar 2022
Dense Passage Retriever - is a set of tools and models for open domain Q&A task.
Role in this project:
Back-end Developer & Data Engineer
Back-end Developer & Data EngineerContributions:4 reviews, 109 commits, 21 PRs in 1 year 10 months
Contributions summary:Vladimir's commits primarily focus on enhancing the data processing and retrieval capabilities within the project. They implemented fixes for file overwriting during data downloads and added training files for the SQUAD dataset. Furthermore, the user integrated multi-task retriever training features, including configuring and adapting the data loading process. They also added the ability to select and use specific tokens for representation selection in the biencoder model.
open-domainretrieverdomainpassagedense
facebookresearch/fairseq
Feb 2019 - Feb 2019
Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
Contributions:2 commits in 1 day
pytorchnlpsequencepythontransformer-architecture
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.