Julian Salazar
Staff Research Scientist at Google DeepMind
San Francisco, California, United States
Summary
👤
Senior🎓
Top SchoolJulian Salazar is a Staff Research Scientist at Google DeepMind with 12 years of experience building generative and real-time speech systems that power multimodal conversational AI. He has led R&D across Gemini, Project Astra, and Lyria RealTime, contributing native audio-out dialog capabilities and publishing high-impact work such as SpeechSSM (ICML'25 oral) and Spectron (ICLR'24). Previously he drove ASR and NLP innovations at AWS AI Labs—work that informed Amazon Transcribe and produced widely cited research like Masked Language Model Scoring and Self-Attention for CTC. Julian blends deep theoretical training (BA in Mathematics, Harvard) with hands-on engineering, from refining LSTM/GRU implementations in popular deep learning texts to open-sourcing long-form speech datasets like LibriSpeech-Long. He also contributes part-time to AI tools for mathematicians, reflecting a rare mix of applied ML, speech/audio expertise, and formal math background. Based in San Francisco, he is known for turning novel research into production-quality models and tooling that scale in real-time applications.
12 years of coding experience
9 years of employment as a software developer
Bachelor of Arts Mathematics, Bachelor of Arts Mathematics at Harvard UniversityHigh School Diploma, High School Diploma at Henry Wise Wood High School
English, Tagalog
Github Skills (10)
mxnet10
machine-learning10
jupyter-notebook10
lstm10
gru10
rnn-model10
deep-learning10
n10
python10
latex9
Programming languages (4)
Github contributions (5)
zackchase/mxnet-the-straight-dope
Aug 2017 - Aug 2017
An interactive book on deep learning. Much easy, so MXNet. Wow. [Straight Dope is growing up] ---> Much of this content has been incorporated into the new Dive into Deep Learning Book available at https://d2l.ai/.
Role in this project:
ML Engineer
ML EngineerContributions:5 commits, 2 PRs, 1 comment in 4 days
Contributions summary:Julian made several commits focused on refining and improving LSTM and GRU implementations within the context of an interactive deep learning book. The changes involved correcting computational errors, enhancing the presentation through better LaTeX formatting, and making minor code adjustments to the existing notebooks. Further contributions include the addition of explanations to the code, clarifying the logic and purpose behind the implementation.
pytorchd2lmxnetdeep-learninghas-content
aws-samples/amazon-transcribe-research-pseudolikelihood
Oct 2019 - May 2020
Contributions:8 commits, 8 pushes in 6 months
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.