Changhan Wang
Technical Lead at Meta
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolChanghan Wang is a Technical Lead and AI researcher with 12 years of experience building production-grade generative and speech models, currently driving speech capabilities for Llama at Meta in New York. He has led core FAIR efforts across speech translation, recognition, and synthesis, and previously implemented foundational sequence-to-sequence models—most notably contributions to the Levenshtein Transformer in Facebook's widely used fairseq toolkit. His background spans both research and production: from deep learning research and graduate instruction at the University of Michigan to machine learning engineering in industry startups, with hands-on expertise in model implementation, services, and deployment. Known for bridging cutting-edge research with scalable engineering, he combines strong academic credentials and published work with practical experience shipping models in large-scale systems.
12 years of coding experience
11 years of employment as a software developer
Fuzhou No.1 Middle SchoolMaster of Science Computer Science, Master of Science Computer Science at University of MichiganVisiting Student Department of Statistics, Visiting Student Department of Statistics at North Carolina State UniversityBachelor of Science (B.S.) Mathematics, Bachelor of Science (B.S.) Mathematics at Zhejiang University
Chinese, English, hokkien
Github Skills (9)
transformer-models10
pytorch10
machine-translation10
neural-machine-translation10
python10
fairseq10
artificial-intelligence9
algorithm9
algorithms9
Programming languages (4)
Github contributions (5)
facebookresearch/fairseq
Sep 2019 - Mar 2022
Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:6 reviews, 77 commits, 14 PRs in 2 years 6 months
Contributions summary:Changhan contributed significantly to the development of the Levenshtein Transformer model, which is a core component of the fairseq toolkit. Their work included implementing the model, task, and criterion classes, as well as adding baseline models like iterative NAT Transformer and CMLM Transformer. They also made changes to the existing codebase, which included bug fixes and addressing dependencies.
pytorchnlpsequencepythontransformer-architecture
facebookresearch/covost
Feb 2020 - Jan 2021
Contributions:8 commits, 9 pushes, 3 comments in 11 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.