I treat my GitHub profile as a living workspace rather than a polished portfolio. Most of what's here is me studying other people's projects โ AI testing platforms, coding-agent skills, and RAG experiments โ by forking them, reading the code, and rebuilding the parts I find interesting from scratch, mostly in Python (crawlers, exercises, and small reproducible scripts).
Real projects I'm reading through, fork by fork โ grouped by what each one teaches me:
| Theme | Project | What I'm learning from it |
|---|---|---|
| ๐ค AI Testing | AITestPlatform | How an LLM tool-calling loop drives the full requirement โ case โ execution โ report flow |
| ๐ค AI Testing | Argus | Turning plain-English test descriptions into plans that actually execute |
| ๐ ๏ธ Agent Skills | codex-ppt-skill | How an image-based deck generator works: prompt โ gpt-image-2 โ packaged slides |
| ๐ ๏ธ Agent Skills | GenericAgent | A self-evolving agent that grows a skill tree from a single seed file |
| ๐ RAG | rag-knowledge-system | Self-hosted RAG: document parsing, chunking strategies, and hybrid retrieval |
๐ These are forks I keep around to read โ click through if you're curious about the same topics.
๐ More forks I'm reading through
| Project | What it does |
|---|---|
| WHartTest | AI-driven test platform: requirement โ executable test cases |
| CLI-Anything | Making software agent-native through CLIs |
| guizang-ppt-skill | HTML slide-deck skill: editorial & Swiss layouts |
- Read the source. Fork it, trace it, then decide what's worth keeping.
- Keep repos small โ small enough to understand in one sitting.
- Write READMEs that don't hide assumptions โ the doc is the design.
- Validate with real commits and pushes, not theoretical reasoning.
- Build the simplest version first, polish only what proves useful.
โญ๏ธ Feel free to look around โ this profile is a workspace, not a showcase.