An AI-powered learning operating system that turns goals into structured study plans. A six-stage LangGraph agent workflow (Profile → Knowledge → Resource → Planner → Critic → Output) uses a Planner-Critic repair loop to keep improving plan completeness and executability. A full RAG pipeline parses PDF/TXT/Markdown, stores vectors in Milvus, and traces citations — supporting material Q&A, summarisation, knowledge-point extraction, and flashcard generation. Includes a PostgreSQL persistence layer plus dashboards for task distribution, daily load, and review progress. Six-layer architecture with a framework-free domain model, and a Fake/Real dual mode toggled by one line of config. 30+ pytest cases.
PythonLangGraphRAGMilvus⭐ 6Projects
About 354 wordsAbout 1 min
A collection of open-source projects I've built and maintained.
Reusable skill modules for learning system design from real codebases. Maps architecture, traces data flow, and coaches design decisions — works with both Claude Code and Codex.
MarkdownA collection of reusable AI-agent skills for scientific research, targeting Nature-style academic writing and figure creation. I contributed the Nature-Paper-Card skill — it generates source-constrained, in-depth Paper Cards (sections 01–16) covering method logic, evidence chains, conclusion boundaries, and critical analysis, helping researchers deeply understand papers and discover new research ideas.
MarkdownForkPersonal blog and portfolio built with VuePress & vuepress-theme-plume, publishing regularly on LLMs, Agents, RAG, AI4Math, computer vision, and AI engineering practice. Also runs the WeChat official account Kstheme for AI (250+ followers).
TypeScriptCopyright
Copyright Ownership:Kstheme
License under:Attribution 4.0 International (CC-BY-4.0)