I specialize in System Programming, High-Performance Computing, and Game/Graphics Engineering. My passion lies in understanding software internals at the binary level and building high-performance tools from scratch. （ソフトウェアの深層理解と、低レイヤー技術を用いた高パフォーマンスなアプリケーションをしています）
情報学部所属。 C++ / DirectX / CUDA を用いたコアシステムの開発を中心に、バイナリ解析技術からGUI/Webアプリケーションもやってます
Core & Low-Level
Engineering & Analysis
DirectX 11 CUDA Win32 API x86/x64 Assembly IDA Ghidra
Application & Web
Runtime Overlay & Process Extension Framework
- Overview: Designed a lightweight internal rendering engine and modding framework for DirectX 11 applications.
- DirectX 11アプリケーション向けの、軽量な内部オーバーレイおよびランタイム拡張エンジンの開発。
- Key Technologies:
- Graphics Pipeline: Custom rendering pipeline hooking via Kiero, implementing UI with ImGui/ImFX.
- Architecture: Implemented EnTT (ECS) for high-performance component management, optimizing memory layout.
- Low-Level Control: Robust API hooking with MinHook and signature scanning with libhat for version-independent compatibility.
- Modern C++: Built with C++17/20, focusing on compile-time optimization and type safety.
V10Inference - Scratch-built AI Inference Engine (Public)
- Overview: Developed a lightweight neural network inference engine from scratch, executing ONNX models on NVIDIA GPUs without relying on high-level frameworks like TensorRT.
- ONNXモデルを解析し、自作のCUDAカーネルを用いて推論を行う軽量エンジンの開発。
- Key Technologies & Implementation:
- Custom CUDA Kernels: Implemented Convolution (Conv2D), MaxPooling, and Softmax kernels manually to optimize parallel execution on GPU.
- Memory Optimization: Managed Host-Device memory transfers (
cudaMalloc,cudaMemcpy) to minimize latency. - No Black Box: Built a custom parser to extract tensor data directly from ONNX Protocol Buffers, strictly understanding the inference logic.
Binary Analysis & Protection Research (C++ / C#)
- Process Instrumentation: Developed a UWP-compatible launcher with memory management and hardware monitoring (using LibreHardwareMonitor).
- Security Mechanisms: Research on software protection techniques, including custom obfuscation algorithms and static analysis mitigation.
- Reverse Engineering: Reconstructing function signatures in stripped binaries to understand internal logic.
- プロセス解析ツール開発および、ソフトウェア保護技術（難読化・解析耐性）の研究・実装。
CppLiveTuner (C++17 / Header-only Library / Public)
- Overview: A cross-platform, header-only library enabling real-time parameter tuning in C++ applications without recompilation.
- 再ビルド待ち時間をゼロにする、C++開発者向けのライブチューニングライブラリ（STBスタイル）。
- Key Technologies:
- Native OS Integration: Utilized ReadDirectoryChangesW (Win) / inotify (Linux) for zero-latency file monitoring.
- Thread Safety: Designed a synchronization mechanism for safe graphics API updates.
StemStudio (Python, C++)
- Overview: AI-based music separation tool combining PyTorch with a C++ backend for performance-critical sections.
- Impact: Developed a hybrid architecture to overcome Python's GIL limitations for real-time processing.
- 音源分離AIを用いたデスクトップアプリ。ボトルネックとなる処理をC++で記述し高速化。
Attendance Reminder (Chrome Extension / JavaScript)
- Problem Solving: Automated attendance tracking helper for students. Published on Chrome Web Store.
- Impact: Solved a real-world problem for university students, improving daily productivity.
- 学生の課題（出席登録漏れ）を解決するChrome拡張機能。ユーザー視点での課題解決。