Fardin Abdi
Principal Scientist at Amazon
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolFardin Abdi is a Principal Scientist and systems-focused ML engineer with 13 years of experience building and productionizing deep learning systems across startups and large tech firms. He has led ML teams at Stripe and Pinterest and now works on foundation models at Amazon’s AGI Labs, blending research rigor from a PhD background with pragmatic engineering. His open-source contributions to high-profile projects like Horovod and Petastorm show a deep understanding of distributed training and data pipeline performance—e.g., adding in-memory caching for PyTorch loaders and modernizing TensorFlow examples. Past work ranges from synthetically generating safe datasets with GANs at Capital One to architecting fault-tolerant software for cyber-physical systems in academia, reflecting a rare combination of safety-critical research and production ML expertise. Colleagues rely on him to marry scalable infra, thoughtful metrics/backtests, and clean code during large migrations and model rollouts.
13 years of coding experience
14 years of employment as a software developer
Bachelor of Science, Electrical Engineering, Bachelor of Science, Electrical Engineering at University of TehranUniversity of Illinois Urbana-Champaign
English, Persian, Turkish, Azerbaijani, Arabic, Kurdish
Stackoverflow
1,324reputation
3.9mreached
22answers
7questions
Github Skills (17)
pytorch10
distributed-training10
python10
machine-learning10
keras10
deep-learning10
tensorflow10
parquet10
pyarrow9
pyspark9
deeplearning-ai9
macos6
neural-network6
lstm6
memory-management6
Programming languages (5)
Github contributions (5)
horovod/horovod
Dec 2018 - Dec 2020
Distributed training framework for TensorFlow, Keras, PyTorch, and Apache MXNet.
Role in this project:
ML Engineer
ML EngineerContributions:11 reviews, 38 commits, 45 PRs in 2 years
Contributions summary:Fardin's contributions primarily involved refactoring and updating existing TensorFlow code within the Horovod framework. Specifically, they removed deprecated TensorFlow dataset APIs and adapted examples, such as `tensorflow_mnist.py` and `tensorflow_mnist_estimator.py`. Furthermore, the user's commits demonstrate an understanding of distributed training environments and integration with Horovod for frameworks such as Keras.
mpikeras-tensorflowtrainingbaidutensorflow
uber/petastorm
Jul 2020 - Dec 2020
Petastorm library enables single machine or distributed training and evaluation of deep learning models from datasets in Apache Parquet format. It supports ML frameworks such as Tensorflow, Pytorch, and PySpark and can be used from pure Python code.
Role in this project:ML Engineer
ML EngineerContributions:1 release, 8 reviews, 6 commits in 5 months
Contributions summary:Fardin contributed to improving the usability of the Petastorm library within the context of PyTorch and TensorFlow. Specifically, the user integrated an in-memory caching mechanism for the PyTorch data loader, enhancing performance. Furthermore, the user exposed pyarrow filters in the make_reader and make_batch_reader APIs, enabling users to filter parquet files at the read level. They also made modifications to examples, and fixed issues related to the library's integration with Apache Spark.
pythontrainingdeep-learning-modelstensorflowdata-science
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.