Roman Levenstein
Engineering Director at Meta
Los Altos, California, United States
Summary
🤩
Rockstar🎓
Top SchoolRoman Levenstein is an Engineering Director based in Los Altos with 14+ years building and leading teams that design ML compilers, runtimes, and software stacks for AI accelerators. He combines deep hands-on expertise in compiler construction (LLVM/GCC), language implementation, and performance optimization with product-scale leadership at Meta, where he leads the compiler and SW stack for MTIA accelerators and was a founding developer of the open-source Glow compiler. Roman holds a strong research-to-production pedigree—over 20 patents and a history of applied telecom research—bringing uncommon fluency across embedded systems, Swift/Java/C++ toolchains, and domain-specific ML languages. He’s equally comfortable optimizing back-end compiler passes (e.g., graph and concat merging in Glow) and shaping cross-functional teams and processes to ship hardware-aware ML software. An active open-source contributor and former Swift performance engineer at Apple, he’s known for marrying deep algorithmic insights with pragmatic engineering to accelerate ML deployment on custom hardware.
14 years of coding experience
22 years of employment as a software developer
Diplom (M.Sc.) Informatik (Computer Science), Diplom (M.Sc.) Informatik (Computer Science) at RWTH Aachen UniversityB.S. Applied Mathematics, B.S. Applied Mathematics at Moscow Aviation Institute (National Research University)
English, German, Russian
Github Skills (17)
c-language10
compile10
compiler-optimization10
cprogramming-language10
jit10
llvm9
performance-optimization9
datastructures-algorithms8
data-structures8
code-generation8
data-structure8
algorithms8
algorithm8
operation7
tensor7
Programming languages (8)
Github contributions (5)
pytorch/glow
Nov 2017 - Oct 2022
Compiler for Neural Network hardware accelerators
Role in this project:
Back-end Developer & Compiler Engineer
Back-end Developer & Compiler EngineerContributions:54 reviews, 422 commits, 251 PRs in 4 years 11 months
Contributions summary:Roman primarily contributed to the development and optimization of the glow compiler for neural network hardware accelerators. Their work included implementing and optimizing compiler passes, such as merging concatenation nodes for graph optimization. They also made code-quality improvements and refactoring, including adding new features for data-parallel operations and integrating external libraries like libjit for efficient code execution.
hardware-acceleratorscompilerneural-networkacceleratorshardware
opti-mix/glow
May 2018 - Apr 2024
Compiler for Neural Network hardware accelerators
Contributions:316 pushes, 90 branches in 6 years
hardware-acceleratorscompilerneural-networkacceleratorshardware
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.