AI Agents From Zero

AI Agents From Zero (AI 智能体实战速成指南) is a free, Chinese-language course for becoming an LLM application or AI agent developer. It combines chapter notes, runnable Python code, full projects and an interview question bank in one GitHub repository, and is updated as the agent ecosystem changes.

What it covers

The outline has six parts:

• LLM foundations: how large models work (Transformer, MoE, attention), running models with Ollama, and prompt engineering.
• Low-code platforms: Coze and Dify workflows, agents and knowledge bases, Docker, deploying models on cloud servers or AutoDL, and AI coding tools such as Trae and Qoder.
• Frameworks: LangChain (LCEL, memory, tools, retrieval, agents), LangGraph (state graphs, persistence, multi-agent), MCP from protocol to custom servers, and the A2A protocol.
• Enterprise RAG and agent projects: multi-route retrieval with vector, sparse and Neo4j search, reranking and RAGAS evaluation.
• Fine-tuning: data preparation, LoRA and QLoRA, LLaMA-Factory, evaluation, deployment and multi-GPU training.
• Engineering practice: how LLM projects are scoped and run inside companies.

The main track is Python with LangChain and LangGraph, deliberately not the Java route of Spring AI or langchain4j.

Projects and interview prep

As of October 2026 the concept chapters are complete, along with two full projects that each have their own source repository: an e-commerce NL2SQL assistant (LangGraph, MySQL, Qdrant, Elasticsearch, FastAPI) and a DeepAgents multi-agent research system with web search, a RAGFlow knowledge base and live progress over WebSocket. More projects appear in the outline. A separate interview bank, organised by skill area and cross-linked to the chapters, collects common questions for AI application developer roles.

How to read it

• Online docs: the whole course as a searchable site.
• Clone the repo to run the examples with Python 3.10 to 3.13. You need an API key from a provider such as Qwen (Alibaba Bailian) or DeepSeek, or you can use a local model through Ollama instead.
• The tutorial itself is in Chinese; an English translation of the README gives an overview.

Who it's for

Chinese-speaking developers, especially frontend, backend or product people, who want a structured path into agent development and something concrete to put on a CV.

License and caveats

MIT, with about 5.4k GitHub stars as of October 2026. The author says the material builds on course notes from 尚硅谷 and other public sources, reworked and extended. The README carries a sponsor block with a referral link for a GPU cloud.

Related on SleekNav

For the opposite approach, an English course that skips frameworks and builds agents by hand in JavaScript with local models, see AI Agents From Scratch.