Guanghan Ning
Member Of Technical Staff at Fleet AI, Inc.
San Francisco Bay Area United States
Summary
👤
Senior🎓
Top SchoolGuanghan Ning is a Member of Technical Staff at Fleet AI with 11 years of experience building and scaling ML systems, currently focused on RL environments and post-training for agentic models. He previously led Code-SFT at ByteDance Seed, bootstrapping coding capabilities and training Dense and MoE models up to 200B parameters, and co-authored the Seed-Coder open release. Earlier tech leadership at JD.COM produced a first-place PoseTrack solution and a production keypoint moderation system that cut manual review costs by 78.8%, and he holds 12 US patents. A PhD in computer vision and a strong open-source footprint — including significant contributions to the YOLO/darknet project and recent puzzle-style ARC environments — give him a rare mix of research depth and production-minded engineering. He brings consistent wins at academic challenges and product impact, with a habit of turning novel data and training recipes into measurable capability gains.
12 years of coding experience
6 years of employment as a software developer
Bachelor of Science (BS), Electrical and Computer Engineering, Bachelor of Science (BS), Electrical and Computer Engineering at Beijing Jiaotong UniversityUniversity of Missouri
English, Chinese
Github Skills (10)
mask-rcnn10
cuda10
computer-vision10
opencv10
faster-rcnn10
c-language10
cprogramming-language10
python9
machine-learning9
object-detection9
Programming languages (4)
Github contributions (5)
Guanghan/darknet
Dec 2015 - Sep 2016
Convolutional Neural Networks
Role in this project:
ML Engineer
ML EngineerContributions:52 commits, 52 pushes in 9 months
Contributions summary:Guanghan significantly contributed to the YOLO object detection project by adding and modifying core functionalities. They implemented a new demo mode enabling video input and output, integrated OpenCV for video processing and display, and updated the yolo.c and yolo_kernels.cu files for video handling. Furthermore, they included a Python script to convert annotation files to a darknet-compatible format. They also modified the class names and other parameters in the `yolo.c` file.
convolutional-neural-networks
Guanghan/mxnet-centernet
Aug 2019 - Nov 2019
Gluon implementation of "Objects as Points", aka "CenterNet"
Contributions:79 commits, 4 pushes in 3 months
centernetdetectorgluonmxnet
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.