Kevin Hu
Technical Staff at Perplexity AI
San Francisco, California, United States
Summary
👤
Senior🎓
Top SchoolKevin Hu is a Technical Staff engineer at Perplexity AI with 13 years of experience building production-grade ML systems, specializing in automatic speech recognition, speech separation, and deep learning. He previously worked on TensorRT at NVIDIA and contributed to TensorFlow's Lingvo project—implementing and refining attention mechanisms and transformer decoder layers to improve speech and translation models. A UC Berkeley EECS graduate with a Public Policy minor, he blends rigorous engineering with an appreciation for broader societal impacts of AI. Based in San Francisco, Kevin moves fluidly between research code and production optimization, with a track record of shipping performant models and runtime enhancements. Notably, his open-source contributions emphasize practical improvements (e.g., batch-major support) that make state-of-the-art models more flexible and scalable.
13 years of coding experience
1 year of employment as a software developer
Bachelor of Science - BS, EECS, Public Policy Minor, Bachelor of Science - BS, EECS, Public Policy Minor at University of California, Berkeley
Chinese, English
Github Skills (13)
attention-mechanism10
transformer-models10
machine-translation10
nlp10
tensorflow10
python10
speech-recognition9
sequence-to-sequence9
speech-synthesis8
accelerated-computing8
distributed-computing8
gpu8
gpgpu8
Programming languages (2)
Github contributions (5)
tensorflow/lingvo
Apr 2020 - Nov 2022
Lingvo
Role in this project:
ML Engineer
ML EngineerContributions:20 commits in 2 years 7 months
Contributions summary:Kevin primarily contributed to the development and testing of machine learning models within the Lingvo framework. Their work focused on implementing and refining attention mechanisms, including multi-source attention and transformer decoder layers. They also made improvements to the existing models, adding support for batch-major and other features. These modifications aimed at enhancing model performance and flexibility for tasks related to speech and translation.
asrtranslationctcspeech-recognitiontensorflow
mrhuke/supervised_two_speaker
Apr 2013 - Jul 2016
Contributions:22 commits, 7 pushes in 3 years 3 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.