AI + Scientific Computing | Agentic learning tools | Research workflow automation
I build tools and workflows that make hard technical work easier to inspect, reproduce, and continue. My current focus is the overlap between AI agents, scientific-computing practice, and documentation-driven engineering.
- Agentic learning systems that use recall, diagnosis, and mastery gates instead of passive chat.
- Scientific-computing workflows for numerical experiments, validation, and reproducible analysis.
- Research automation that connects code, evidence, review loops, and durable documentation.
- AI agents, LLM workflows, prompt/process design
- Python automation, validation scripts, data-processing tools
- CUDA/GPU scientific computing and numerical simulation workflows
- Git, shell, Markdown, and documentation-heavy engineering
- Reproducible validation, audit trails, and handoff-ready project structure
| Project | What it is | What it demonstrates |
|---|---|---|
ContextCanopy |
A local-first personal AI substrate that gives multiple agents shared memory, identity, goals, rules, and evidence. | Multi-Agent continuity, Markdown knowledge graphs, MCP, evidence governance, transactional writes, and deterministic validation. |
tutor |
A file-based AI tutor template for mastery learning, active recall, and spaced repetition. | Agent workflow design, stateful learning systems, validation scripts, privacy boundaries, and reusable project structure. |
More public research and tooling projects will be added here as they are cleaned, documented, and safe to share.
I like systems where the important state is visible, the assumptions are written down, and the next person can verify what happened without guessing. That usually means small scripts, clear status files, reproducible checks, and documentation that is part of the engineering surface rather than an afterthought.
Email: zhangyswx@163.com