Kazuhiro Serizawa
Software Engineer, AI LLM Engineering Team at Mercari, Inc.
Chiyoda, Japan
Summary
🤩
Rockstar🎓
Top SchoolKazuhiro Serizawa is a seasoned software engineer with 12 years of experience who blends web backend craftsmanship with deep ML and high-performance inference engineering. Currently on an ML & Data Ops team in Tokyo, he has driven production LLM inference optimizations (including vLLM and Triton integrations), built Redis-backed rate limiters and real-time inference pipelines, and tuned throughput with speculative decoding. His background spans Ruby, Go, Python and Rust, plus significant work on distributed systems and orchestration across AWS/GCP, Kubernetes and workflow engines like Digdag (to which he has contributed). He has an academic publication on overlapping data staging for ML I/O and a track record of improving robustness in open-source libraries such as google-auth-library-ruby. Comfortable moving between low-level performance tuning and higher-level platform design, he often finds practical trade-offs that unlock faster ML serving in production.
12 years of coding experience
14 years of employment as a software developer
Master of Engineering - MEng, Computer Science, Master of Engineering - MEng, Computer Science at 筑波大学Bachelor's degree, Agriculture, General, Bachelor's degree, Agriculture, General at 岩手大学 / Iwate University
Japanese, English
Github Skills (12)
javas10
error-handling10
dag10
faraday10
java10
ruby10
databases9
database9
database-api9
rspec9
testing8
devops8
Programming languages (8)
Github contributions (5)
googleapis/google-auth-library-ruby
Dec 2016 - Jun 2018
Google Auth Library for Ruby
Role in this project:
Back-end Developer & DevOps Engineer
Back-end Developer & DevOps EngineerContributions:11 commits, 3 PRs, 11 comments in 1 year 6 months
Contributions summary:Kazuhiro primarily focused on improving the robustness and stability of the Google Auth Library for Ruby. Their contributions include implementing retry mechanisms for network-related errors in key components like token revocation and access token retrieval. They also addressed code style issues through automated rubocop corrections and improved error handling. Further work involved fixing potential edge cases and refactoring to use existing retry mechanisms, improving the resilience of the library.
auth-librarygoogle-auth-libraryrailsrubygoogle-auth
treasure-data/digdag
Jan 2018 - Sep 2018
Workload Automation System
Role in this project:Back-end Developer
Back-end DeveloperContributions:68 reviews, 82 commits, 50 PRs in 8 months
Contributions summary:Kazuhiro primarily contributed to the Digdag project by implementing and modifying back-end functionalities. Their work included adding dependencies for copied tasks, implementing a TDResultExportOperatorFactory and adding tests. This involved changes to database interaction and integrating with Treasure Data, showcasing their focus on extending the system's data processing capabilities. They also modified the UI to show logs by task, which demonstrates a concern for improving user experience.
workload-automationdigdagsystem-automationautomationworkload
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.