Peter Buchlovsky
Software Engineer at Google DeepMind
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolPeter Buchlovsky is a software engineer based in San Francisco specializing in speeding up large-model training across TensorFlow, JAX, and TPU compiler stacks at Google DeepMind. With a background in high-throughput distributed systems at Goldman Sachs and machine vision/ML at HP, he brings a rare mix of production-scale systems engineering and low-level compiler/accelerator work. He is an active open-source contributor to flagship projects like JAX and TensorFlow/XLA, adding performance paths, GPU lowering rules, and tooling to dump and analyze latency-hiding schedules. Peter holds a First Class BSc and a Distinction MSc in Computer Science from the University of Birmingham and has pursued research at Cambridge, reflecting strong academic foundations. Colleagues rely on him for pragmatic optimizations that bridge research-grade ML code and production accelerator toolchains.
7 years of coding experience
6 years of employment as a software developer
Research Student, Computer Science, Research Student, Computer Science at University of CambridgeBSc, Artificial Intelligence and Computer Science, First Class Honours, BSc, Artificial Intelligence and Computer Science, First Class Honours at University of Birmingham
Github Skills (23)
c-language10
artificial-neural-networks10
optimizers10
python10
machine-learning10
optimizer10
numpy10
tensorflow10
neural-network10
compiler10
jax10
cprogramming-language10
protobuffer10
protobuf10
debug9
Programming languages (4)
Github contributions (5)
google-deepmind/sonnet
May 2019 - Oct 2019
TensorFlow-based neural network library
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:1 review, 47 commits in 5 months
Contributions summary:Peter primarily focused on refactoring and improving the Sonnet library's optimization modules. Their contributions involved moving common optimizer utility functions and adding support for IndexedSlices to several optimizers, including SGD, Momentum, RMSProp, and Adam. They also added the TpuReplicator to the distribute API, implying work on improving the libraries' distributed training capabilities. The user made several code modifications to support the library's TensorFlow-based machine learning goals.
deep-learningneural-networksmachine-learningneural-networktensorflow
openxla/xla
Nov 2019 - Jan 2023
A machine learning compiler for GPUs, CPUs, and ML accelerators
Role in this project:Backend Developer
Backend DeveloperContributions:6 commits in 3 years 2 months
Contributions summary:Peter primarily contributed to the XLA compiler's backend, implementing and modifying features related to scheduling and debugging. They added functionality to dump the latency-hiding schedule to a protobuf format, including module and core frequency information, for debugging and analysis. Furthermore, the user made a small modification, likely a nit, to code related to latency estimation within the latency hiding scheduler. They also added a function for building DebugOptions flags.
compilercommunity-drivenmachine-learningmodular
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.