didilili/ai-agents-from-zero

What is this project?

ai-agents-from-zero (Chinese name: 《AI 智能体实战速成指南：从零到企业级落地》) is a free, open-source, Chinese-language tutorial for learning how to build AI agents — software that uses large language models (LLMs) to perform tasks. The project's stated goal is to be a complete learning path from absolute beginner to enterprise-level deployment, covering tutorials, runnable code, interview prep, and real projects.

It positions itself as a systematic alternative to scattered blog posts and paid bootcamps, and is continuously updated (targeting 2026).

What's inside

The repository contains four main components:

1. Systematic tutorial chapters — organized into 6 sections:

   • LLM basics: model architecture (Transformer, MoE, self-attention), prompt engineering, deployment platforms (Ollama, cloud/local)
   • Low-code platforms: Coze (扣子) and Dify, plus Docker and enterprise deployment
   • Core development frameworks: LangChain, LangGraph, MCP protocol, A2A protocol
   • Enterprise RAG/Agent projects: five real projects (see below)
   • Fine-tuning: LoRA/QLoRA, LLaMA-Factory, memory optimization, multi-GPU training
   • Big-company engineering standards: research process, project design, evaluation

2. Runnable example code — every case is meant to actually run, with environment setup and troubleshooting notes

3. An interview question bank — organized by job skill areas, with questions drawn from real big-company interviews and public experience reports

4. Two completed full projects (with separate source repos):

   • 电商问数 (E-commerce Query): a natural-language-to-SQL system using MySQL, LangGraph, Qdrant vector search, Elasticsearch, and FastAPI — completed May 3
   • 深度研搜 (Deep Research Search): a multi-agent research system built on DeepAgents, combining web search, MySQL, RAGFlow knowledge base, file generation, and real-time progress via WebSocket — completed May 17

Three more projects are outlined (掌柜智库, 电商小二, 市场罗盘) but not marked as complete.

Tech stack focus

The tutorial deliberately focuses on the Python ecosystem — specifically Python + LangChain + LangGraph — and explicitly avoids the Java route (Spring AI, langchain4j). Other covered technologies include: Coze/Dify low-code platforms, vector databases, MCP/A2A protocols, Docker/Ollama/vLLM for deployment, and PEFT/LoRA/QLoRA for fine-tuning.

Who is it for?

• Beginners with zero background (concepts are explained from the ground up)
• Working engineers who want to deepen their skills
• People transitioning from frontend/backend/product roles into AI development
• Job seekers targeting roles like AI Application Engineer, AI Agent Engineer, or AI Product Technical Lead

Current status

• Concept chapters: fully updated
• Two practical projects: completed (May 2026)
• The outline is explicitly described as not fixed — new topics and projects will be added as the AI ecosystem evolves

How to get started

1. Clone the repo: git clone https://github.com/didilili/ai-agents-from-zero.git
2. Create a Python virtual environment (Python 3.10 recommended, supports 3.10–3.13) and install dependencies
3. Copy .env-example to .env and fill in an API key (e.g., Alibaba Qwen, DeepSeek)
4. Run the first example script from the project root
5. Optionally use a local Ollama model to avoid needing a cloud API key

The README also links to an online readable version of the docs.

Community and sponsorship

• A WeChat community group called 「胡思妙想研究所」 ("Absurd Ideas Research Institute") — join by commenting on GitHub issue #46
• The project is sponsored by 优云智算 (Compshare), a UCloud AI cloud platform offering GPU plans starting at 49 RMB/month; new users via the referral link get 5 RMB of free trial credit

Key caveats

• The tutorial references materials from a paid course (尚硅谷《大模型智能体速成班》) but has been restructured and expanded with public docs and community practice
• The interview question bank includes questions from real big-company interviews but has been reorganized around the tutorial's own chapter structure
• All content is in Chinese (an English translation of the README is linked but the main content is Chinese)