Orhan Firat
Research Scientist at Google DeepMind
New York, New York, United States
Summary
👤
SeniorOrhan Firat is a research scientist based in New York with 12 years of experience advancing large language model research, development, and deployment across top AI labs. He has held research roles at Google, Google DeepMind, Meta, IBM, and Mila, bridging deep academic training with production-focused ML engineering. Orhan’s contributions to prominent open-source projects like TensorFlow Lingvo and Theano/Blocks reveal hands-on work in Transformer optimization, stability fixes for sequence models, and robust testing infrastructure. He combines a strong background in sequence modeling and attention mechanisms with practical expertise in optimizers (AdamW) and activation improvements (GELU), helping move models from research prototypes toward scalable systems. Known for both research leadership and code-level rigor, he often surfaces subtle stability and testing improvements that reduce production risk. Based in NYC, he blends academic depth with industry impact, making him effective at translating cutting-edge models into reliable deployments.
11 years of coding experience
8 years of employment as a software developer
Github Skills (20)
unit-testing10
machine-translation10
python10
testing10
machine-learning10
recurrent-neural-networks10
ml10
mle10
transformer-models10
tensorflow10
neural-network10
nlp10
theano10
distributed-computing9
attention-mechanism9
Programming languages (1)
Github contributions (5)
mila-iqia/blocks
Mar 2015 - Aug 2015
A Theano framework for building and training neural networks
Role in this project:
ML Engineer
ML EngineerContributions:21 commits, 5 PRs, 13 comments in 5 months
Contributions summary:Orhan primarily contributed to the `blocks` Theano framework, focusing on enhancements related to recurrent neural networks and sequence generation. Their work included refactoring cost calculation methods for sequence generators, adding tests for the cost function, and addressing stability issues in attention mechanisms. The commits demonstrate an understanding of core components within the library and also involve modifications related to auxiliary variables.
pytorchdeep-learningtheanoneural-networksmachine-learning
tensorflow/lingvo
Aug 2018 - Apr 2019
Lingvo
Role in this project:ML Engineer
ML EngineerContributions:7 commits, 2 comments in 7 months
Contributions summary:Orhan's contributions primarily involve modifications and additions to the Lingvo framework, a machine learning library. Their work includes moving machine translation models, adding learning rate decay functionality to Transformer models, and incorporating configurations for Transformer-Base models. They have also added GELU non-linearity, AdamW optimizer, and modifications to TransformerAttentionLayer. These changes suggest a focus on model development and optimization within the Lingvo ecosystem.
asrtranslationctcspeech-recognitiontensorflow
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.