Pedram Pejman
Staff Engineer at Google DeepMind
New York, New York, United States
Summary
👤
Senior🎓
Top SchoolPedram Pejman is a Staff Engineer in New York with 11 years building production ML infrastructure and training platforms across Google DeepMind and Google product teams. He designs and ships model adaptation systems—recently focused on reinforcement learning and fine-tuning workflows for Gemini and previously Imagen—while also contributing to widely used open-source tooling like TensorFlow Serving. Pedram blends hands-on engineering (prototyping new tuning products and RL training pipelines) with program leadership, having launched internal standards and scaled teams rapidly during high-impact rollouts. His background spans distributed systems, Kubernetes, cloud AI programs and product work at Google Play and Azure, giving him a rare perspective across research, infra and productization. Unconventional for an engineer of his profile, he wrote an undergraduate thesis proposing a unified, agent-agnostic intelligence evaluation framework and remains active creatively as a musician.
11 years of coding experience
9 years of employment as a software developer
Thomas Jefferson Highschool for Science and TechnologyB.S. in Commerce B.S. in Computer Science, B.S. in Commerce B.S. in Computer Science at University of Virginia
Github Skills (12)
continuous-deployment10
machine-learning10
deep-learning10
tensorflow10
serve10
python10
ml-deployment10
apidoc9
deep-neural-networks9
neural-network9
api9
cpp8
Programming languages (7)
Github contributions (5)
tensorflow/serving
Apr 2019 - Aug 2020
A flexible, high-performance serving system for machine learning models
Role in this project:
ML Engineer
ML EngineerContributions:14 commits, 6 PRs, 104 comments in 1 year 4 months
Contributions summary:Pedram contributed significantly to the TensorFlow Serving project by enhancing its capabilities. Their work included adding tests for loading SavedModels created with the Keras Sequential API, which improves model serving compatibility. They also integrated the project with the TensorBoard profiler service, allowing for performance analysis and debugging. Furthermore, they introduced the ability to poll the filesystem for updated model configurations and added support for version labels in the HTTP REST API, improving model deployment flexibility and management.
cpppythonservingdeep-learningml
peddybeats/peddy.ai
Mar 2019 - Nov 2024
Personal blog content and structure
Contributions:2 PRs, 100 pushes, 3 branches in 5 years 9 months
jekyll
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.