Kai Tai
Research Scientist at Meta
Castro Valley, California, United States
Summary
🤩
Rockstar🎓
Top SchoolKai Tai is a research scientist with 11 years of experience building and deploying machine learning systems, currently leading development of Meta’s internal text embedding service and on-device LLM efforts. His background spans deep learning research and production engineering—from PhD work at Stanford to shipping multi-GPU Torch libraries and medical and large-scale vision models at MetaMind/Salesforce. He has hands-on expertise in efficient inference, sparse model training, and implementing core model architectures (including TreeLSTM and neural style transfer enhancements) reflected in notable open-source contributions. Based in Castro Valley, Kai combines rigorous academic training in physics and computer science with a pragmatic focus on performance and deployability, often surfacing subtle implementation fixes that improve real-world model robustness.
11 years of coding experience
7 years of employment as a software developer
Master of Science (MS), Computer Science, Master of Science (MS), Computer Science at Stanford UniversityBachelor of Arts (BA), Physics, magna cum laude, Bachelor of Arts (BA), Physics, magna cum laude at Princeton University
Github Skills (15)
neural-style-transfer10
sentiment-analysis10
lua10
algorithms10
pytorch10
machine-learning10
computer-vision10
lstm10
nlp10
deep-learning10
natural-language-processing9
cudnn9
data-structures8
algorithm8
data-structure8
Programming languages (4)
Github contributions (5)
kaishengtai/neuralart
Aug 2015 - Mar 2017
An implementation of the paper 'A Neural Algorithm of Artistic Style'.
Role in this project:
ML Engineer
ML EngineerContributions:22 commits, 5 PRs, 26 pushes in 1 year 6 months
Contributions summary:Kai primarily focused on enhancing a neural style transfer implementation. Their contributions included adding command-line options for image size, display control, and optimization algorithms. They integrated total variation smoothing for better image quality and handled monochrome image inputs. The user also added support for both Inception and VGG network models, significantly expanding the project's capabilities.
pytorchdeep-learningartisticneural-algorithmmachine-learning
stanfordnlp/treelstm
Mar 2015 - Jul 2017
Tree-structured Long Short-Term Memory networks (http://arxiv.org/abs/1503.00075)
Role in this project:ML Engineer
ML EngineerContributions:18 commits, 1 PR, 16 pushes in 2 years 4 months
Contributions summary:Kai implemented LSTM baselines for semantic relatedness and sentiment analysis tasks within the tree-structured LSTM framework. They added new modules and functionalities for LSTM-based models, including both single and bidirectional LSTM architectures, and integrated them into existing training and prediction pipelines. Furthermore, they addressed and corrected bugs related to model structure and data handling during the training process.
memoryarxivabsstructureddeep-learning
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.