✓ Checked
Learn AI, LLMs and Agents: 10 GitHub Repos From a Viral List
A viral post lists 10 GitHub repos for learning Python, ML, LLMs and AI agents. We checked each: course facts, licenses, last activity, a link that now leads to a renamed product, two research repos superseded or frozen, and a learning order.
On October 21, 2026 Rakib Hossen (@rakib_hossen_ai) called ten GitHub repositories a "goldmine for AI engineers": a path from Python basics through machine learning and LLMs to AI agents and production apps, ending with the advice to pick one, build something, then move on.
We opened each repository: its README, its license and when it was last changed. The four Microsoft courses, the LLM book code, the OpenAI Cookbook and the Python course are alive and match what the post says about them, with one caveat on language. Three links need a warning. Two are research releases from 2022 and 2023 that have not been pushed to since 2024, and for one of them the same team now points to newer models. And the LLM App link no longer leads to LLM App: the repository was renamed and is now a different product.
The original post by @rakib_hossen_ai on X (October 21, 2026)
The ten, checked
| Repo | What it is | License | Last push | Status |
|---|---|---|---|---|
| jackfrued/Python-100-Days | 100-day Python course, written in Chinese | none stated | Jul 29, 2026 | Active; one main author |
| microsoft/ML-For-Beginners | 12 weeks, 26 lessons, 52 quizzes of classic machine learning | MIT | Oct 21, 2026 | Active |
| microsoft/AI-For-Beginners | 12 weeks, 24 lessons: neural networks, vision, language | MIT | Oct 21, 2026 | Active |
| rasbt/LLMs-from-scratch | Code for the book Build a Large Language Model (From Scratch) | Apache 2.0 (GitHub shows NOASSERTION) | Oct 2, 2026 | Active |
| microsoft/generative-ai-for-beginners | 21 lessons on building generative AI apps | MIT | Oct 22, 2026 | Active |
| microsoft/ai-agents-for-beginners | 18 lessons on building AI agents | MIT | Oct 21, 2026 | Active |
| openai/openai-cookbook | Examples and guides for the OpenAI API | MIT | Oct 23, 2026 | Active |
| pathwaycom/llm-app | Now redirects to pathwaycom/serviette, a no-code RAG server | MIT, plus a free Pathway license key | Oct 16, 2026 | Renamed; templates moved |
| CompVis/stable-diffusion | Original Stable Diffusion v1 research code | CreativeML OpenRAIL-M | Jun 18, 2024 | Not pushed to since 2024; 620 open issues |
| facebookresearch/segment-anything | The first Segment Anything Model (SAM) | Apache 2.0 | Sep 18, 2024 | README points to SAM 2; SAM 3 is out |
Licenses, dates and issue counts from GitHub's API and each repository's README and license file, read October 23, 2026. None of the ten is archived.
The course numbers in the post hold up
The post calls ML-For-Beginners a "12-week, 26-lesson" curriculum and AI-For-Beginners "12-week, 24-lesson". Both match the repositories' own descriptions: "12 weeks, 26 lessons, 52 quizzes, classic Machine Learning for all" and "12 Weeks, 24 Lessons, AI for All!". The post gives no lesson counts for the other two Microsoft courses: Generative AI for Beginners has 21 lessons, and its README says the code comes in both Python and TypeScript; AI Agents for Beginners has 18.
The topic lists are also fair. Generative AI for Beginners has lessons on prompt engineering, RAG and vector databases, AI agents and fine-tuning, as the post says. AI Agents for Beginners covers tool use, agentic RAG, multi-agent design, protocols such as MCP and A2A, memory and security. Two things the post does not mention: the agent course's code uses the Microsoft Agent Framework with Microsoft Foundry Agent Service, and its README marks Microsoft Foundry as "Azure Account Required", so plan for an Azure account. The Generative AI course notes that GitHub Models, used in some of its lessons, was to retire at the end of July 2026, and points to Microsoft Foundry Models instead.
All four Microsoft courses are translated into many languages; the README of the agent course warns that its 50+ translations make the download much larger and shows how to clone without them.
Two research repositories that stopped moving
Segment Anything. The post presents facebookresearch/segment-anything as the place to explore computer vision. The repository still works as a record of the first SAM, under Apache 2.0, but its last push was on September 18, 2024 and it has 596 open issues. Its own README opens with "Latest updates -- SAM 2" and sends readers to the new release for images and video. Meta has since gone further: the facebookresearch/sam3 repository holds SAM 3, Segment Anything with Concepts, and its README reports SAM 3.1 checkpoints released on March 27, 2026, which it says are significantly faster for tracking many objects (as reported). Two differences matter before you switch: SAM 3 is under its own "SAM License" rather than Apache 2.0, and you must request access to its checkpoints on Hugging Face first.
Stable Diffusion. CompVis/stable-diffusion is the original v1 release by the research group that built it with Stability AI and Runway. Its last push was on June 18, 2024 and it has 620 open issues. The weights are under the CreativeML OpenRAIL-M license, which allows commercial use but adds use-based restrictions, and the README itself says the weights "are research artifacts and should be treated as such". It is good reading on how latent diffusion works. To generate images in code, the README itself shows the Diffusers library as a simple way to download and sample the model.
LLM App is now a different product
The post's link, github.com/pathwaycom/llm-app, now redirects to pathwaycom/serviette. GitHub describes it as "a no-code RAG server that keeps answers in sync with your documents". The README opens with a note for people looking for the old llm-app templates: all of them, with the cookbooks and examples, now live in the Pathway repository under examples/llm-app.
So the stars the post points to (58,779) belong to a repository whose main product changed. Serviette itself is MIT-licensed, but it runs on the Pathway framework, which needs a free license key (the README links a one-click sign-up with GitHub or LinkedIn), and its quick demo uses an OpenAI API key. Per its README, serviette answers one question at a time, from the current state of the documents; conversation memory, user accounts and a hosted offering are not built in yet. If you want the RAG and real-time pipeline examples the post describes, go to the Pathway repository's examples folder.
Python - 100 Days is written in Chinese, with no license
The most-starred repository on the list (187,386 stars) is a Chinese-language course; its title on GitHub is "Python - 100天从新手到大师". The post does not mention the language. It covers far more than the post's one line suggests: Python basics, files and Office documents, databases, Django, web scraping, data analysis and some machine learning, mostly written by one author (457 commits by jackfrued on the first page of contributors). GitHub reports no license and the repository has no license file, so anything beyond reading it needs the author's permission. If you do not read Chinese, it is the wrong first step.
LLMs from scratch is a book's companion code
rasbt/LLMs-from-scratch is the official code repository for Sebastian Raschka's book Build a Large Language Model (From Scratch). The notebooks follow the book's chapters, from text data and attention to pretraining and two kinds of fine-tuning, and the README says the main chapters are designed to run on conventional laptops without special hardware (as reported). You can work through the code without the book, but the explanations are in the book. GitHub cannot name the license (NOASSERTION); the LICENSE.txt file is the Apache License 2.0.
Stars and recent interest
Total stars rank the list differently: Python-100-Days leads with 187,386, then Generative AI for Beginners (121,955) and LLMs-from-scratch (107,129). Recent stars show where learners go now. The agent course and the AI course gained the most over those six months. Most of AI-For-Beginners' gain came at once: 12,443 new stars in August 2026, 53% of its six-month total by our calculation, with 2,305 on August 1 alone. AI Agents for Beginners grew more steadily, with at least 2,294 new stars in every month from April to September. The two research repositories and the renamed llm-app gained between 1,290 and 1,605 each in the same period.
A learning order that works
The post's closing advice, pick one and build something, is sound. Here is an order that follows the courses' own prerequisites. The agent course itself says that if this is your first time building with generative AI models, take Generative AI for Beginners first.
If your goal is to build AI agents rather than research models, you can go straight from Python to Generative AI for Beginners and come back to steps 2 and 3 later. A few of our pages fit along the way:
- Steps 2 and 3: our pages on scikit-learn, PyTorch, TensorFlow and Jupyter show how widely each library is used.
- Before the agent course: What Is an AI Agent? and What Is MCP?, which lesson 11 of the course builds on.
- After it: How to Build an AI Agent and Best AI Agent Frameworks compare the Microsoft Agent Framework the course uses with the OpenAI Agents SDK, LangGraph and others; the Cookbook's examples use the OpenAI Python library.
What the post gets right
- Seven of the ten are the right repositories for the job the post gives them, and all seven had a push in 2026.
- The lesson counts it gives are correct, and the topic lists match the courses.
- The advice to pick one path and build fits how the Microsoft courses are set up: each lesson can be taken on its own, and the agent course sends newcomers to the generative AI course first.
What it leaves out
- The llm-app link redirects to a renamed repository, now a no-code RAG server; the templates the post describes moved to the Pathway repository.
- Segment Anything has been followed by SAM 2 and SAM 3 from the same team; the original repository has not been pushed to since September 2024.
- The Stable Diffusion repository is v1 research code, not pushed to since June 2024, with weights under a license with use restrictions.
- Python-100-Days is in Chinese and has no license.
- The agent course needs an Azure account for its main code samples, and LLMs-from-scratch is the code for a paid book.
Where to start
- New to programming: a Python course in your own language, then ML-For-Beginners.
- You code and want to build with LLMs: Generative AI for Beginners, then AI Agents for Beginners, then the OpenAI Cookbook.
- You want to know how LLMs work inside: LLMs-from-scratch, ideally with the book.
- Computer vision: start from SAM 3 or SAM 2 rather than the original Segment Anything, and read the SAM License first.
- Image generation: read the Stable Diffusion repository for the ideas; use Diffusers to run models.
We did not run these courses or models; everything above comes from the ten repositories, the SAM 3 repository, their README and license files, GitHub's API and AgentGid's pages, read October 23, 2026.