Gang Liao
Research Scientist at Meta
Mountain View, California, United States
Summary
🤩
Rockstar🎓
Top SchoolGang Liao is a research scientist with 11 years of experience building ML and data infrastructure from silicon to serving stacks, currently driving monetization-level ML serving at Meta in Mountain View. He has deep expertise in scalable storage and columnar systems—authoring and leading projects such as a next-generation column store for ML and contributing performance and blob-caching improvements to the widely used RocksDB. His background spans both industry research and production engineering across TikTok, Microsoft Research, Baidu, and ByteDance, with publications on HTAP and integer decoding for fast analytics. Gang combines systems-level optimizations (CPU/GPU-aware algorithms and LSM-level blob extraction) with ML-serving practice for large LLMs, enabling remote, high-throughput ranking models. He is notable for bridging academic rigor from a PhD program with hands-on open-source impact on projects that power real-world ML workloads.
11 years of coding experience
5 years of employment as a software developer
Doctor of Philosophy - PhD Computer Science, Doctor of Philosophy - PhD Computer Science at University of Maryland
Chinese, English
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (21)
caching10
storage-engine10
c-language10
admin10
python10
image-classification10
rocksdb10
databases10
blob-storage10
database-administration10
performance-optimization10
paddlepaddle10
computer-vision10
cprogramming-language10
database10
Programming languages (13)
Github contributions (5)
PaddlePaddle/book
Mar 2017 - Sep 2017
Deep Learning 101 with PaddlePaddle （『飞桨』深度学习框架入门教程）
Role in this project:
ML Engineer
ML EngineerContributions:88 commits, 57 PRs, 66 pushes in 6 months
Contributions summary:Gang contributed to the image classification tasks within the PaddlePaddle book repository. Their work involved adding and updating API versions and model definitions, specifically focusing on image classification using the PaddlePaddle framework. The commits demonstrate the implementation of ResNet and VGG models for image classification tasks and updating related documentation and training scripts.
deep-learningpytorchpaddlepaddletensorflow
facebook/rocksdb
May 2022 - Aug 2022
A library that provides an embeddable, persistent key-value store for fast storage.
Role in this project:Back-end Developer
Back-end DeveloperContributions:296 reviews, 19 commits, 21 PRs in 2 months
Contributions summary:Gang primarily focused on enhancing the RocksDB storage engine. Their contributions included optimizing the handling of blob files during database open, specifically integrating file sizes from the manifest. Additionally, the user implemented features to support more granular garbage collection policies for blobs within the database, allowing for compacting specific ranges and overriding existing configurations. Further work involved enabling the extraction of large values into blob files, starting from a specified LSM tree level, and integrating blob caching to improve performance.
persistent-storagebigtablefast-storagelsm-treedatabase
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.