Aaron Sarna
Senior Software Engineer at Google
Cambridge, Massachusetts, United States
Summary
🤩
Rockstar🎓
Top SchoolAaron Sarna is a Senior Software Engineer based in Cambridge, MA with six years of professional experience blending web development, browser extensions, and machine learning research. Currently at Google, he applies practical ML engineering to computer vision and NLP, contributing notable improvements to high-profile Google Research projects like SupCon and the Scenic JAX library (adding optimizers, fixing activations, and improving checkpoint resumption). His background includes shipping popular cross-browser extensions (Forecastfox) and low-latency systems work from internships in HFT and web infrastructure, giving him a rare mix of front-end UX, systems performance, and ML model-tuning expertise. Comfortable across the stack, he focuses on reproducible training, optimizer and hyperparameter work, and browser-native UX. Collected studies in CS, cognitive science, and Near Eastern studies hint at a broad intellectual curiosity that informs both product thinking and research-driven engineering.
6 years of coding experience
3 years of employment as a software developer
BA, Computer Science, Cognitive Studies, Near Eastern Studies, BA, Computer Science, Cognitive Studies, Near Eastern Studies at Cornell Universityhigh school diploma, high school diploma at Maimonides School
Hebrew, Arabic, Spanish
Github Skills (14)
transformers10
hyperparameter-tuning10
computer-vision10
machine-learning10
deeplearning-ai10
deep-learning10
jax10
tensorflow10
resnet10
python10
contrastive-learning10
imagenet10
mlops9
attention-mechanism8
Programming languages (5)
Github contributions (5)
google-research/scenic
Mar 2022 - Jul 2022
Scenic: A Jax Library for Computer Vision Research and Beyond
Role in this project:
ML Engineer
ML EngineerContributions:5 commits in 3 months
Contributions summary:Aaron primarily contributed to the core machine learning aspects of the Scenic library. Their work included adding a LARS optimizer, correcting a missing GELU activation in a BERT model layer, and modifying the training utils. They also made changes relating to how a job restarts and the global step is handled during resuming model training from a checkpoint, ensuring correct randomization.
scenictransformersvision-transformervisiondeep-learning
google-research/google-research
Oct 2020 - Nov 2020
Google Research
Role in this project:ML Engineer
ML EngineerContributions:5 commits in 1 month
Contributions summary:Aaron contributed significantly to the SupCon (Supervised Contrastive Learning) project within the Google Research repository. Their work focused on fine-tuning hyperparameters, including learning rates, weight decay, and optimizer parameters (RMSProp and LARS), across multiple ResNet architectures (ResNet50, ResNet101, ResNet200) for the ImageNet dataset. The user also addressed software compatibility and bug fixes as well as updating and adapting loss functions.
googlemachine-learningai
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.