Boncheol Gu
Chief Software Architect at FuriosaAI
Gyeonggi, South Korea
Summary
👤
Senior🎓
Top SchoolBoncheol Gu is Chief Software Architect at FuriosaAI with over a decade of experience building compilers, simulators, and runtime libraries for neural network accelerators. He combines deep embedded systems expertise from a long tenure at Samsung—where he designed SSD in-storage computing and hierarchical storage software—with hands-on accelerator toolchain development. His open-source contributions include bug fixes and correctness improvements to pytorch/glow, linking his work to a prominent compiler for neural network hardware. A Ph.D. researcher trained at Seoul National University, he brings rigorous systems thinking to production-focused machine learning infrastructure. Based in Gyeonggi, South Korea, Boncheol blends low-level performance engineering with higher-level compiler and runtime design, making him comfortable across silicon-to-software stacks. He often surfaces subtle correctness and performance fixes—such as allocator elimination and node naming improvements—that improve long-term maintainability of complex graph compilers.
10 years of coding experience
4 years of employment as a software developer
Ph.D. Computer Science and Engineering, Ph.D. Computer Science and Engineering at Seoul National UniversityBachelor Computer Science, Bachelor Computer Science at Yonsei University
Korean, English
Github Skills (12)
pytorch10
compiler-design10
c-language10
deep-learning10
cprogramming-language10
tensorflow9
operation9
tensorrt9
tensor9
graph-algorithms8
testing8
optimization7
Programming languages (3)
Github contributions (5)
pytorch/glow
Jun 2018 - Aug 2018
Compiler for Neural Network hardware accelerators
Role in this project:
Backend Engineer
Backend EngineerContributions:5 commits, 4 PRs, 8 comments in 2 months
Contributions summary:Boncheol contributed to the `pytorch/glow` repository, a compiler for neural network hardware accelerators, by fixing bugs and improving the codebase. Their contributions included addressing an issue related to the `group` parameter within the `Interpreter::fwdConvolutionGradInst` function. They also fixed variable erasure in the graph and corrected an assertion related to extractSlice operations. Finally, the user fixed deleteDeadAllocs and made changes to use `std::string` as a return type of `Node::getInputName()`.
compilerhardwareneural-network
furiosa-ai/torch-fx-rs
Nov 2023 - Sep 2025
Contributions:5 reviews, 3 PRs, 5 pushes in 1 year 9 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.