Ayush Dubey
Software Engineer at Google DeepMind
Mountain View, California, United States
Summary
🤩
Rockstar🎓
Top SchoolAyush Dubey is a software engineer with 13 years of experience building high-performance distributed systems and ML infrastructure, currently working on Gemini Infra and ML Pathways at Google DeepMind in Mountain View. He has deep expertise in TensorFlow runtime and distributed training—contributing to core runtime abstractions, networking, and multi-worker strategies—and has improved documentation and tutorials for the wider TensorFlow community. His background combines academic rigor (PhD from Cornell) with practical systems work at Google, where he’s driven hardware-software co-design, memory tiering, and performance optimization across fleet-scale systems. Notably, he introduced abstractions like DistributedContext and FabricCommunicator to enable scalable distributed execution, and has hands-on experience resolving tricky consistency and checkpointing issues in multi-worker environments.
13 years of coding experience
6 years of employment as a software developer
Indian Institute of Technology Delhi (IIT Delhi)Doctor of Philosophy (Ph.D.) Computer Science, Doctor of Philosophy (Ph.D.) Computer Science at Cornell University
Github Skills (17)
c-language10
learn-ruby-on-rails10
python10
user-manual10
machine-learning10
distributed-systems10
basics10
deeplearning-ai10
deep-learning10
tensorflow10
resnet10
cprogramming-language10
documentation10
async9
asynchronous9
Programming languages (4)
Github contributions (5)
tensorflow/models
Feb 2019 - Jan 2020
Models and examples built with TensorFlow
Role in this project:
MLOps Engineer
MLOps EngineerContributions:22 commits, 23 PRs, 8 pushes in 11 months
Contributions summary:Ayush primarily focused on integrating and optimizing distributed training strategies for ResNet models within the TensorFlow framework. Their contributions include implementing multi-worker support, migrating from `CollectiveAllReduceStrategy` to `MultiWorkerMirroredStrategy`, and configuring cluster specifications. The user also addressed checkpointing behavior within the multi-worker environment, and introduced command-line options. They updated distribution strategy flags and optimized input data sharding for distributed training, improving performance and scalability.
deep-learningtensorflow
tensorflow/runtime
May 2020 - Dec 2020
A performant and modular runtime for TensorFlow
Role in this project:Back-end Developer
Back-end DeveloperContributions:11 commits in 7 months
Contributions summary:Ayush primarily contributed to the TensorFlow Runtime project by implementing and refining core functionalities. They made consistent improvements to the `DenseHostTensor` operations, ensuring consistency between operation definitions and kernel definitions, likely to improve user experience. Further, they introduced crucial abstractions like `DistributedContext` and `FabricCommunicator` to facilitate distributed runtime environments, supporting network communication and data transfer. They also addressed bugs and refactored code to improve software quality, as well as introducing a `CallbackRegistry` to support asynchronous point-to-point data transfers.
runtimeperformantmodulartensorflow
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.