English / 简体中文
PhD student at the Southern University of Science and Technology (SUSTech), working across scientific computing, quantum network and control engineering. Interested in AI4S research.
hadamard is a family of tools built around LLM harnesses. Hadamard Canvas (Hcanvas) (0.1.x) is a local-first, AI-native canvas for engineering and research work — one document model shared by humans and agents, shipped as a desktop app plus a small CLI/HTTP surface that agents can read and patch directly, with export to HTML, PDF, PNG, PPTX, and DOCX. Hadamard Harness is the experimental agent runtime in the same family: a TypeScript runtime with durable sessions, tools, skills, memory, MCP, and a bridge to installed agent CLIs (Claude Code, Codex, and more).
- 📖 Docs: https://deconbear.github.io/hadamard/
- 📦 npm:
hadamard-canvas - 🛠️ TypeScript · Electron desktop · CLI/HTTP agent protocol · MCP · Bridge
- 🎨 HTML / canvas / infinite / Markdown pages · 66 PPT-style shapes · local-first, no cloud account
SmithTool is an interactive Smith Chart desktop application for RF impedance matching and matching-network design, built with C++17 and Qt 6 Widgets. It offers interactive chart visualization, source/load impedance editing (rectangular and polar), series/shunt R-L-C matching element design, matching wizards for L/Pi/T and other topologies, Touchstone import, and SPICE netlist export.
- 🛠️ C++17 · Qt 6 Widgets · CMake
- 📡 Touchstone (
.s1p/.s2p) import · SPICE export · English / Simplified Chinese UI
notebook (🧪 Beta) is a diagram-driven learning repository — one folder per chapter — that breaks down how AI works and how to build it, covering AI, control theory, robotics, world models, information theory, quantum information, algorithms, and ROS 2. Each article focuses on a single concept readable in 20–30 minutes, and all code runs on CPU by default, so a consumer laptop is enough.
- 📖 Read online: https://deconbear.github.io/notebook/
- 🛠️ Python · PyTorch · illustrated notes + runnable code
- 🎯 From perceptrons to LLMs: backpropagation, attention, RLHF, RAG, Bellman equations, and more
causalwave is a causal numerical solver for linearly coupled propagation equations — no training or GPU required. marcher solves one-way spatial propagation combined with local time evolution, IF2 provides spatial Volterra elimination and second-order exponential time stepping for the two-level GEM model, and an optional XMDS2 interface covers input generation, isolated builds, and result reading, with derivations and reproducible validation included.
- 📖 Docs: https://deconbear.github.io/causalwave/
- 🛠️ Python · NumPy / SciPy · GEM / Maxwell–Bloch · XMDS2 interop
- 🎯 Numerical baseline for quantum-memory GEM simulations
xmds-extended is an unofficial derivative of XMDS2 focused on practical cross-platform maintenance, with native Windows support as its first major downstream addition — no MSYS or WSL required. It ships a portable Windows installer with upgrade and uninstall flows, CI for Windows, Linux, and macOS, and bilingual project pages and tutorials.
- 📖 Project site: https://deconbear.github.io/xmds-extended/
- 🛠️ Python · C++ · XMDS2 · Cross-platform CI



