I'm a Machine Learning Engineer and undergraduate researcher at the Takenawa Laboratory, Tokyo University of Marine Science and Technology. I enjoy turning research ideas into reliable software, especially where search algorithms, machine learning, and backend systems meet.
- Research: Monte Carlo Tree Search (MCTS), reinforcement learning, GBDTs, and efficient search
- Engineering: Go, Python, CMake, PostgreSQL, Flutter, and cloud infrastructure
- Current focus: search-budget optimization and lightweight learning systems
wang-bin/fvp — merged
PR #381: Support checksum-pinned MDK SDK dependencies
- Added optional SHA-256 verification for CMake dependency downloads to make native builds reproducible.
- Implemented cache invalidation, atomic replacement, concurrency protection, and recovery from interrupted installs.
- Added network-free CMake tests for Ubuntu and Windows and improved the related CI and Android example build.
rlglab/minizero — merged
PR #13: Expose MiniZero environments through Python bindings
- Added a pybind11 interface that lets Python research code directly use MiniZero's compiled C++ game environments.
- Exposed environment control, legal actions, rewards, feature tensors, action history, and metadata with safe NumPy ownership and invalid-action handling.
- Validated the API across TicTacToe, Go, and 2048, including configuration-dependent board sizes and environment-specific reset signatures.
PR #12: Replace non-standard bitset _Find_first usage
- Replaced the non-standard
std::bitset::_Find_first()dependency with a portable helper while preserving the existing libstdc++ fast path. - Added a Clang/libc++ fallback that scans 64-bit chunks with
__builtin_ctzll(), improving macOS toolchain compatibility. - Updated call sites across the Go, Havannah, and KillallGo environments and validated the fallback and MiniZero Python module builds.
| Project | What it demonstrates |
|---|---|
| Pet Platform API | A deployed Go/PostgreSQL API with Firebase authentication, SQL migrations, Docker, and documented curl examples. |
| original_LLM | A small Japanese decoder-only Transformer trained from scratch in PyTorch without pretrained models or fine-tuning. |
| Ramen Radar | A Flutter application that combines Google ratings and real travel distance to rank nearby ramen shops. |
- Machine learning: MCTS, AlphaZero-style systems, reinforcement learning, deep learning, and GBDTs
- Backend: Go, Python, Django REST Framework, PostgreSQL, REST APIs, and Docker
- Client: Flutter, Riverpod, Next.js, and TypeScript
- Infrastructure: AWS, Render, Supabase, GitHub Actions, and CMake
- Undergraduate Researcher, Takenawa Laboratory
- 4th-year undergraduate, Tokyo University of Marine Science and Technology (TUMSAT)
- Major: Logistics and Information Engineering
I'm open to conversations about ML research, open-source collaboration, and engineering opportunities.
- Website: suzukioff.com
- LinkedIn: Natsuhiro Suzuki
- Email: suzuki@suzukioff.com