Build AI from scratch: a 503-lesson Chinese course with no copy-paste

What it does

This is the Chinese-language derivative of Rohit Ghumare’s open-source AI Engineering from Scratch curriculum. It spans 503 lessons across 20 phases—roughly 320 hours—taking learners from linear algebra and calculus all the way to autonomous agent clusters and MCP servers, with implementations in Python, TypeScript, Rust, and Julia. Every lesson follows the same loop: derive the math, build the algorithm from scratch with zero dependencies, then rebuild it with production frameworks so you understand what the abstraction is actually doing.

The interesting bit

The localization is not a lazy machine-translation mirror. It runs a standalone Chinese website with interactive charts, dark mode, and progress tracking; bakes in 3Blue1Brown-style animated explainers with Chinese dubbing; and ships CI checks to ensure the lesson count never drifts out of sync with the actual content. Most unusually, every single lesson ends with a tangible, reusable artifact—prompts, agent skills, or MCP servers—that you can install directly into Claude, Cursor, or any compatible client.

Key highlights

• 503 lessons across 20 phases, from math foundations to multi-agent systems
• Build-it-first pedagogy: hand-write algorithms in raw code before touching PyTorch or sklearn
• Fully localized Chinese website (aieng-zh.cn) with search, command palette, and interactive diagrams
• Each lesson ships reusable outputs: prompts, SKILL.md files, agents, and MCP servers
• AI-retrieval optimized with auto-generated llms.txt, sitemaps, and structured data

Caveats

• Animated explainer videos are only finished for Phase 1 (22 lessons); remaining phases are still in production
• At ~320 hours, this is explicitly pitched as the opposite of a quick-start tutorial

Verdict

Worth bookmarking if you are a Chinese-speaking developer who wants to stop treating LLMs like black boxes and start building them from the ground up. Skip it if you are looking for a quick API-tuning cheat sheet or if you do not read Chinese—the localization layer is the whole point.