China's AI Agent Education Ecosystem Goes Systematic: The ai-agents-from-zero Phenomenon

Full English Translation

The Most Systematic AI Agent Crash Course on the Internet (From Zero to Enterprise Deployment)

Continuously updated through 2026 · Goal: the “strongest on Earth” AI Agent tutorial — systematic course + runnable source code + interview question bank + enterprise-level hands-on projects, fully aligned with “AI Agent / LLM Application Development Engineer” training syllabi and job descriptions, a complete learning path.

Tutorial Highlights

• Systematic learning path, one line through: From LLMs and prompts, to low-code (Coze/Dify) and code frameworks (LangChain/LangGraph), to enterprise RAG/Agent, fine-tuning, and engineering standards — organized by knowledge system in a complete closed loop, suitable for systematic mastery rather than fragmented bookmarking.
• Python ecosystem focus, avoiding Java detours: Many agent courses center on Spring AI and langchain4j, leaning toward the Java stack; this tutorial’s main thread is Python only, with emphasis on LangChain / LangGraph, making it a true LLM agent crash course.
• Accessible, genuinely suitable for zero-foundation beginners: The entire tutorial is arranged from shallow to deep — first clarifying core concepts like LLMs, prompts, Agents, RAG, and MCP — then progressively entering frameworks and hands-on projects, minimizing opaque jargon for first-time learners.
• Enterprise-level practice, aligned with “ability to deliver”: Uses merchant operations assistants, e-commerce Q&A, deep research, shopkeeper AI brain, intelligent customer service, and market compass as threads, threading together intent parsing, multi-source knowledge, human handoff, review and monitoring — placing multi-recall, evaluation, observability, cost, and guardrails into real-world contexts for practice.
• Every case meets the “runnable” standard: Not just concepts or pseudocode — provides runnable examples, source code, environment instructions, and common troubleshooting. Cases are organized by actual hands-on experience to help you avoid pitfalls.
• Tutorial-source-interview, three-in-one that actually works: Beyond “reading,” you can “run” and “answer” — runnable cases and source code, prompt templates, and deployment strategies, rejecting “concepts only.” The interview question bank unpacks questions by competency domains from training programs and job descriptions, with substantial portions sourced from real big-tech interviews, public interview reports, and high-frequency follow-up scenarios.

Agent Project Architecture

The repository’s architecture diagram shows an intelligent agent ecosystem covering: multi-model APIs, MCP protocol, LangChain/LangGraph orchestration, Coze/Dify low-code platforms, vector databases, knowledge graphs, RAG pipelines, observability, and deployment.

What You Gain After Completing

• Deployable project capability: independently deliver AI Agent applications (from environment to deployment), advancing from “only calling APIs” to practical engineering.
• Systematized architectural expression: able to clearly articulate RAG, Agent, and MCP design decisions and trade-offs, standing up to scrutiny in interviews and resumes.
• Interview and job description alignment: standalone interview question bank interlinked with chapter content, organized by job competency domains with question formats and answer approaches.
• Engineering and resume material: enterprise-oriented cases with multi-recall, observability, cost, and other formulations — projects are demonstrable and resume-worthy.
• Searchable knowledge map: systematic table of contents + case source code, aligned with common “agent/application development” course dimensions.

Technology Stack Overview

Category	Technologies
LLM Fundamentals	LLM, Transformer, MoE, Self-Attention, LLaMA/Qwen/GPT, Multimodal, Pre-training/Fine-tuning/Inference
Prompting & Orchestration	Prompt Engineering, Chain of Thought, Few-shot, Multi-turn Dialogue, Memory Management, Agent & Tool Invocation
Low-Code Platforms	Coze, Dify — Workflows, Agents, Knowledge Bases, Plugins, Python Integration
Development Frameworks	LangChain, LangGraph — Model I/O, Chains, Memory, Agents (ReAct), Retrieval, Graph-based Workflows
Protocols & Communication	MCP (Model Context Protocol), A2A — Function Calling, Service Decoupling, Cross-Agent Communication
RAG & Retrieval	Vector Databases, Sparse Retrieval, Neo4j, HyDE, BGE-Rerank — Multi-recall, Knowledge Graphs, RAGAS Evaluation
Document & Multimodal	MinerU, OCR — Mixed-format PDF parsing, equipment manuals, after-sales guides
Deployment & Operations	Docker, Ollama, Xinference, vLLM — Tencent Cloud/Alibaba Cloud, AutoDL, Local Deployment
Fine-tuning & Training	PEFT, LoRA, QLoRA, DeepSpeed, Llama-Factory — Alpaca/ShareGPT data formats, Safetensors/ONNX
Programming & Tools	Python; Trae AI, Qoder (AI-assisted IDEs) — multi-model APIs, MCP integration and debugging
Job Hunting & Interviews	Interview question bank — organized by competency domains with question format + answer approach

Tutorial Outline (Excerpt)

01 LLM Fundamentals

• LLM understanding and environment setup
• LLM architecture principles (Transformer, MoE, Self-Attention)
• LLM orchestration platforms (Ollama, cloud/local deployment)
• Prompt engineering (core principles, CoT, Few-shot, multi-turn dialogue)

02 Enterprise Low-Code Platform Development

• Coze platform (UI, plugins, knowledge bases, workflows, agents)
• Project 1: Merchant Operations Assistant
• Dify AI platform (workflows, agents, knowledge bases, multi-case)
• Containerization (Docker)
• Enterprise LLM deployment (Tencent Cloud, Alibaba Cloud, AutoDL)
• AI coding tools: Trae AI, Qoder

03 Core Development Frameworks

• LangChain principles and applications
• LangGraph principles and applications
• MCP from principles to practice
• Cross-Agent communication: A2A protocol

04 Enterprise RAG / Agent Projects

• Shopkeeper AI Brain (LangGraph RAG workflows, MinerU/OCR, multi-recall, RAGAS evaluation)
• Intelligent Customer Service (intent parsing, multi-source knowledge bases, human handoff)
• E-commerce NL2SQL (MySQL, LangGraph, Qdrant, Elasticsearch, FastAPI SSE)
• Deep Research (DeepAgents multi-agent, web search, MySQL, RAGFlow, WebSocket real-time progress)

05 LLM Fine-tuning Practice

• Fine-tuning core concepts
• Enterprise fine-tuning dataset construction
• Efficient fine-tuning with Llama-Factory
• Optimization cases

06 Enterprise Development Standards

• Enterprise LLM R&D processes
• Current LLM hotspots