Sho Watanabe
software
developer
About
Graduate student researcher and software engineer specialized in high-performance systems, reinforcement learning, control theory, and AI agents.
Education
- Bachelor of Engineering — Mathematical Engineering and Information PhysicsGPA: 3.79 / 4.3
Conducted research on online control of linear dynamical systems under uncertainty.
- Analysis, Algebra, Probability
- Optimization, Algorithms
- Electromagnetism, Quantum Mechanics
- Control Theory, Signal Processing
- Computer Architecture
- Master of Information Science and Technology (expected) — Information Physics and ComputingGPA: A (Current)
Member of Computing Systems Research Laboratory.
Skills
Languages
- Rust
- C++
- Python
- CUDA
- C
- R
ML Skills
- JAX
- PyTorch
- TensorRT
- NEAT
- Reinforcement Learning
Tools
- Linux (Arch)
- Docker
- GitHub Actions
- Make
- Ray
Cloud & Infra
- AWS
- Terraform
- CI/CD Pipelines
Software Engineering
- HPC / JAX XLA optimization
- Distributed RL rollout system
- High-performance systems design (Rust)
- Performance optimization (JAX, TensorRT)
- Cross-language integration (Rust ↔ Python FFI)
- Distributed training pipelines (Ray)
- Research-oriented development
- Rust/Python library on PyPI
Projects
High-performance RL simulator written in Rust and JAX.
- Probabilistic dynamics
- Sparse, dynamic action space
- Rust + CUDA hybrid
- Memory-efficient rollouts
- Rust
- Python
- CUDA
NEAT-inspired, optimization-agnostic topology evolution library.
- Rust + PyO3
- Python API for JAX/NumPy/PyTorch
- Automated PyPI builds
- Rust
- Python
- PyO3
- maturin
LLM-powered multi-agent system with strict JSON-schema reasoning.
- Multi-agent LLM workflow
- Tree of Thoughts multi-step refinement
- Strict JSON-schema guided reasoning
- Ollama backend local LLM
- Python
- Ollama
High-performance control theory simulator in Rust and JAX.
- Online control algorithm evaluation
- Hybrid Rust–Python integration
- Rust
- JAX
- Python
AI photo context rediscovery app using BLIP-2 + LLaMA2.
- TensorRT acceleration
- Multimodal reasoning
- Python
- TensorRT
Reproducible, GPU-enabled Docker environment for running modern Shogi AI engines with a GUI.
- GPU-enabled Docker environment
- Plug-and-play Shogi AI practice
- User-friendly GUI setup
- Docker
- Make
Awards
The University of Tokyo
Awarded for research in Control Theory and Online Machine Learning.
Asahi Newspaper
Achieved 1st place out of 16 participants by combining fine-tuned RoBERTa and LightGBM to identify viral news articles using multimodal inputs including images, texts, and metadata.
Signate
Earned a Silver Medal (0.983 mAP50-95) among 123 participants using YOLO11.
Signate
Achieved a Bronze Medal (0.9980 F1) among 338 participants using fine-tuned ViT (EVA).
AtCoder Competitive Programming
OngoingAtCoder
Green coder in weekly contests, using Rust and C++ to strengthen algorithmic thinking and implementation skills.
Experience
- August 2025 – October 2025
Software Engineering Intern @ Woven by Toyota
Tokyo, Japan - On-siteSummary
- Developed multi-object tracking pipelines and deployed production.
Achievements
- Multi-object tracking with Kalman smoother
- AWS + Terraform deployment by automated CI/CD pipelines
- Git-driven collaborative engineering workflow
- Python
- AWS
- Terraform
- Docker
Education
- Bachelor of Engineering — Mathematical Engineering and Information PhysicsGPA: 3.79 / 4.3
Conducted research on online control of linear dynamical systems under uncertainty.
- Analysis, Algebra, Probability
- Optimization, Algorithms
- Electromagnetism, Quantum Mechanics
- Control Theory, Signal Processing
- Computer Architecture
- Master of Information Science and Technology (expected) — Information Physics and ComputingGPA: A (Current)
Member of Computing Systems Research Laboratory.
Skills
Languages
- Rust
- C++
- Python
- CUDA
- C
- R
ML Skills
- JAX
- PyTorch
- TensorRT
- NEAT
- Reinforcement Learning
Tools
- Linux (Arch)
- Docker
- GitHub Actions
- Make
- Ray
Cloud & Infra
- AWS
- Terraform
- CI/CD Pipelines
Software Engineering
- HPC / JAX XLA optimization
- Distributed RL rollout system
- High-performance systems design (Rust)
- Performance optimization (JAX, TensorRT)
- Cross-language integration (Rust ↔ Python FFI)
- Distributed training pipelines (Ray)
- Research-oriented development
- Rust/Python library on PyPI
Awards
The University of Tokyo
Awarded for research in Control Theory and Online Machine Learning.
Asahi Newspaper
Achieved 1st place out of 16 participants by combining fine-tuned RoBERTa and LightGBM to identify viral news articles using multimodal inputs including images, texts, and metadata.
Signate
Earned a Silver Medal (0.983 mAP50-95) among 123 participants using YOLO11.
Signate
Achieved a Bronze Medal (0.9980 F1) among 338 participants using fine-tuned ViT (EVA).
AtCoder Competitive Programming
OngoingAtCoder
Green coder in weekly contests, using Rust and C++ to strengthen algorithmic thinking and implementation skills.
Experience
- August 2025 – October 2025
Software Engineering Intern @ Woven by Toyota
Tokyo, Japan - On-siteSummary
- Developed multi-object tracking pipelines and deployed production.
Achievements
- Multi-object tracking with Kalman smoother
- AWS + Terraform deployment by automated CI/CD pipelines
- Git-driven collaborative engineering workflow
- Python
- AWS
- Terraform
- Docker