Philipp Moritz
Berkeley, California, United States
Summary
🤩
RockstarPhilipp Moritz is a co-founder and CTO based in Berkeley with 16 years of experience building scalable AI and distributed systems. He pairs a PhD in computer science from UC Berkeley with an academic grounding in physics and mathematics, bringing research rigor to production engineering. His open-source work spans high-throughput ML inference and serving (vLLM—adding fused Mixture-of-Experts support, FP8 optimizations, and LoRA infrastructure), distributed runtimes (low-level Ray object store, serialization, and parallel transfers), and data interchange (Apache Arrow). Philipp is known for tackling low-level memory and performance challenges that make large-model serving and distributed ML practical at scale. He focuses on turning cutting-edge ML research into reliable, production-grade infrastructure.
16 years of coding experience
Github Skills (57)
mixtures10
pytorch10
c-language10
caffe10
python10
testing10
memory-management10
arrowkeys10
machine-learning10
data-serialization10
object-storage10
inference10
cmake10
data-structure10
reinforcement-learning10
Programming languages (18)
Github contributions (5)
amplab/SparkNet
Nov 2015 - Apr 2016
Distributed Neural Networks for Spark
Role in this project:
ML Engineer
ML EngineerContributions:135 commits, 55 PRs, 156 pushes in 5 months
Contributions summary:Philipp's contributions primarily focused on integrating and extending the Caffe deep learning framework within the SparkNet project. They added a Java data layer, which enables the loading of data from Java, likely facilitating the integration of Spark RDDs as input to the neural networks. The user also implemented tests for Cifar10 datasets, expanded the Caffe library and provided image preprocessing capabilities. Furthermore, the user contributed to the creation of a C wrapper for Caffe.
neural-networksmachine-learningsparkscaladistributed
ray-project/tutorial
Jul 2017 - Oct 2019
Role in this project:ML Engineer
ML EngineerContributions:13 commits, 63 PRs, 48 pushes in 2 years 2 months
Contributions summary:Philipp implemented a policy gradient algorithm for training a CartPole environment, indicating a focus on reinforcement learning. The user's code modifications involve setting up and training a policy network using TensorFlow and the gym environment. They focused on implementing the core training loop, including action selection, reward calculation, and model updates. Additionally, the user worked on parallelizing the policy gradient implementation using Ray.
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.