A Chinese-curated AI syllabus built for LLM ingestion

What it does

AI-Compass is a curated, Chinese-language knowledge repository that maps AI concepts, tools, and trends across nine modules—from LLM theory and RAG architectures to reinforcement learning and enterprise frameworks like Baidu Paddle and Huawei MindSpore. It publishes weekly highlight digests tracking new models and tools, and includes a small collection of runnable code demos including RAG examples. The repository is explicitly organized to be forked and queried by AI coding assistants such as Claude Code and Codex as a local knowledge base.

The interesting bit

Rather than treating GitHub as a billboard, the authors designed the repo as ingestible context: they recommend feeding specific files like weeklyHighlights/latest.md and weeklyHighlights/INDEX.md to agents so they can distinguish long-term curated content from weekly incremental updates. It is essentially a syllabus formatted for LLM consumption.

Key highlights

• Nine content modules covering theory, frameworks (training, inference, RLHF), applications (RAG, Agent, GraphRAG, MCP), and enterprise open-source tools
• Weekly digests (27+ issues tracked) summarizing new AI models, products, and research releases
• Explicitly structured for AI assistant ingestion with recommended entry points and query patterns
• Includes runnable code demos, though currently limited mainly to RAG samples
• Heavy integration with Chinese tech ecosystems (WeChat, Zhihu, CSDN, Gitee, Coze) and domestic enterprise frameworks

Caveats

• The repository is primarily curated documentation and links rather than original software; the code module is currently thin
• Much of the social ecosystem (WeChat groups, Zhihu columns, knowledge planet coupons) is inaccessible without Chinese-language accounts and platforms
• The README promises ongoing expansion across multiple modules, so some sections appear to be placeholders or works in progress

Verdict

Developers who want a structured, Chinese-language AI curriculum that can be dropped into Claude Code or similar tools for contextual Q&A will find this useful. If you are looking for a standalone framework or extensive original codebase to import, this is not that.