AI Dragon Taming Notes: A Full-Stack AI Engineering Knowledge Repository

What ai_wiki Is and Why It Exists

Most AI engineering knowledge is scattered: blog posts that go stale, papers that skip implementation details, and documentation that assumes a specific framework version. The ai_wiki repository, titled AI Dragon Taming Notes in the README, takes a different approach. It is an open collection of engineering notes organized by topic, each grounded in practical problem-solving rather than theoretical exposition.

The repository is maintained by charliedream1 (Yi Li) and has been updated continuously since 2022. The README's stated goal is to share "problem-solving strategies and key points from engineering practice, share various practical cases, track frontier technology developments, and encompass AI full-stack knowledge." The primary language listed by GitHub is Jupyter Notebook, reflecting that many of the most technical sections include runnable code cells alongside explanatory text.

This is not a course or a structured learning path. The README's quick-start guidance is explicit: select a topic from the directory listing, read the content, apply it to a real problem. The value is in the depth per topic and the breadth of coverage, not in a curated progression.

How the Repository Is Organized

The repository uses a numbered directory scheme with descriptive names. The structure groups content into six broad areas. Programming basics and engineering practice covers directories 01 through 04: system platforms, code snippets and templates (Python, C/C++, Shell, PowerShell), algorithm fundamentals, and object storage.

Database and storage knowledge covers directories 05 through 07: relational databases (MySQL, Redis), vector databases (Milvus, Chroma, Faiss, Vespa, ElasticSearch), and graph databases (Neo4j, NebulaGraph).

Machine learning and deep learning theory covers directories 10 through 15: mathematics foundations, supervised and unsupervised ML, neural network theory, reinforcement learning, graph neural networks, and data engineering.

Traditional deep learning applications covers directories 20 through 24: image recognition (object detection, OCR, document layout analysis), NLP text processing (vector models, reranking, text chunking, knowledge graphs), audio (speech recognition, speaker verification, TTS, audio LLMs), time series, and video analysis.

LLM applications covers directories 31 through 39: LLM training and inference, multimodal models, image and video generation, prompt engineering, RAG (retrieval-augmented generation), agents, and other applications including speech bots and AI scraping. Tools and deployment covers directories 38 through 42.

A final group covers condensed notes (50 and 51), career topics, job resources, interview preparation, and soft skills (directories 60 through 78).

Using the Repository: Clone and Navigate

There is no package to install and no environment to configure to use this repository. The primary usage pattern is cloning and browsing locally or reading directly on GitHub.

Clone the repository to get all content locally:

git clone https://github.com/charliedream1/ai_wiki

This downloads the full directory tree. After cloning, navigate to any numbered directory such as 31_LLM/ or 35_RAG/ and open the Markdown files or Jupyter notebooks relevant to your topic. A Gitee mirror at gitee.com/charlie1/ai_wiki.git is available for readers in regions where GitHub access is slow or restricted.

The README lists a citation in BibTeX format for academic attribution, attributed to Yi Li (charliedream1), published on GitHub in 2022. Community discussion is available through GitHub Discussions and issues. The README also references a paid knowledge community (Zhihu Xiaozhishu) and a WeChat public account for supplementary content, but these are external to the repository.

Jupyter notebooks in the technical sections can be run locally if you have the relevant Python environments set up, but the repository does not include a requirements file or a setup script. Each notebook or code file stands alone with its imports; you are expected to resolve dependencies based on what the specific file requires.

Coverage Depth: LLM and RAG Sections

The LLM section (31_LLM/) is the most extensive based on its place in the README's knowledge map. It covers LLM principles, training data considerations, model training, inference acceleration, service deployment, evaluation, hallucination handling, and safety. This is not a single document: the directory likely contains dozens of files organized by subtopic.

The RAG section (35_RAG/) documents retrieval-augmented generation across several sub-areas: principle overviews, document parsing, retrieval optimization, knowledge graph RAG, memory RAG, Agentic RAG, internet RAG, and evaluation. Each sub-area reflects a real engineering concern that practitioners encounter when building RAG pipelines.

The agent section (36_Agent/) covers agent principles and key components, training, evaluation and tracking, framework usage, commercial case studies, observability, and deployment practice. This kind of engineering-focused coverage of agents, grounded in real deployment scenarios, is harder to find in formal documentation.

The README notes the NLP text processing section (21_NLP/) covers vector models, reranking, text chunking, knowledge graphs, text classification, and clustering. These are all sub-problems that appear when building LLM-adjacent systems, and having engineering-focused notes on each in a single repository is a practical advantage over searching for them separately.

Limitations and What This Repository Is Not

The repository does not carry a stated license in the GitHub metadata. The README does not include a LICENSE file among the top-level entries listed. This matters for engineers who want to adapt the content: without a clear license, the default copyright applies, meaning reuse beyond personal reference is technically restricted. For internal team wikis or reference materials, this is generally not a concern. For republication or use in commercial products, it warrants a direct question to the maintainer.

The content is written primarily in Simplified Chinese. Engineers who do not read Chinese will find the repository less useful even when the code samples are language-agnostic. The README notes a README_EN.md exists for the top level, but per-directory content is in Chinese throughout.

The quality varies by section. Some directories are detailed, with worked examples and code. Others may contain only brief notes or links. The README provides a high-level description of each section's scope, but the only way to evaluate a specific area is to browse it.

LangChain's documentation and the Hugging Face course are common alternatives for structured LLM learning. The difference is that ai_wiki is not a curriculum: it does not guide you through prerequisites. It also lacks the automated testing and version management of a formal documentation system. Notes that were accurate for a specific library version may not reflect current APIs without the maintainer updating them.

Related Project and Community

The README mentions a related repository, ai_quant_trade, described as a quantitative trading platform covering stock knowledge, strategy examples, factor mining, machine learning, deep learning, reinforcement learning, and related areas. Both repositories are maintained by the same author. The ai_quant_trade repository is hosted on GitHub at charliedream1/ai_quant_trade and mirrored on Gitee.

The ai_wiki repository is maintained as a solo project with occasional community contributions through issues and discussions. The README includes a donation section, suggesting the project is sustained by individual maintainer effort rather than organizational funding.

Sections 70 through 78 cover work-related content including career planning, job listings, interview preparation, soft skills, and team management. This is unusual for a technical repository and reflects the author's intent to document not just engineering knowledge but the surrounding professional context. Engineers looking only for technical content can ignore these directories.

Editorial conclusion

ai_wiki suits engineers who want a reference alongside active project work, not a tutorial sequence. The structure rewards direct lookup: pick a directory, open a file, apply it. There are no installation steps, no dependencies, and no build process. The absence of a stated license is a practical concern for anyone who wants to republish or adapt the content in a commercial context. The repository's last push was on 2026-09-20, and the README cites the original publication year as 2022, indicating a long-running project with continued updates.

Frequently asked questions

Official sources

1. charliedream1/ai_wiki on GitHub
2. Issues
3. README