bishwaghimire/ai-learning-roadmaps: a Markdown index for AI self-study
A complete, structured hub for learning Artificial Intelligence — covering AI, Machine Learning, Deep Learning, and Data Science with books, roadmaps, and curated resources from beginner to advanced.
At a glance
- What is it?
- This repository is a curated set of Markdown roadmaps and resource tables covering AI, ML, DL, data science, LLMs, RAG, MLOps, research and AI safety. It is a reading list with a directory structure, not software you install, and its value depends entirely on whether the linked material is still current.
- Who is it for?
- Adopt this if you want a single directory of Markdown roadmaps to work through in order, and you are willing to check each linked book, course or paper yourself before committing months to it. Do not adopt it if you need graded exercises, a runnable environment or a maintained dependency you can pin.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 93 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What the repository actually is
There is no code here. The repository is a collection of Markdown files, one per roadmap, plus tables of books, courses, playlists and papers. The README describes it as a curated hub for Computer Science students, AI enthusiasts and professionals who want to move from beginner to advanced levels. The topics list names ai-for-beginners, ai-learning, ai-roadmap, artificial-intelligence, computer-vision, data-science, deep-learning, llm, machine-learning, nlp, rag and roadmap, which matches the file layout described in the README.
The problem it addresses is navigation, not computation. Anyone starting in machine learning meets an oversupply of material and no ordering. This repository answers the ordering question with a set of paths through named subjects. It does not answer the question of whether you have understood anything, because there are no exercises, no quizzes and no reference implementations in the material provided.
The roadmap files and how they are grouped
The README splits the roadmaps into four bands. Foundations holds ai-roadmap.md, data-science-roadmap.md, ml-roadmap.md and deep-learning-roadmap.md. Specialization holds computer-vision-roadmap.md, nlp-roadmap.md, llm-roadmap.md, generative-ai-roadmap.md and rag-roadmap.md. Engineering and production holds a single file, mlops-production-ai-roadmap.md. Research, safety and long-term AI holds research-scientist-roadmap.md and ai-safety-alignment-roadmap.md.
Each entry in the README carries a one-line gloss. The LLM roadmap is described as covering pretraining, fine-tuning, alignment and evaluation. The RAG roadmap is described as covering vector search, embeddings, system design and evaluation. The MLOps roadmap is described as covering deployment, monitoring, scalability and reliability. Those glosses are the only detail available about the contents of the individual files; the supplied README does not reproduce them.
A career table maps ten goals to sequences. An AI Engineer is pointed at AI, then ML, then DL, then CV or NLP, then LLM. A GenAI Engineer is pointed at AI, DL, LLM, GenAI, RAG. The README states plainly that these are guidelines and not strict rules, and that the roadmaps are modular but connected.
The setup table is the only runnable part
Getting Started is a five-row table, and it is the closest thing to an executable instruction in the repository. It lists Python 3.10+, VS Code, venv, notebooks (Google Colab or Jupyter), and a link to a local file, Packages.md, described as Essential Python Libraries for AI/ML.
That is the whole installation story. There is no requirements.txt, no environment.yml and no lockfile mentioned anywhere in the supplied material, so the dependency versions are whatever the linked libraries publish on the day you install them. Packages.md is a Markdown file of library names, not a pinned set. If you want reproducibility, you will have to build it yourself from that list.
The README does not give a command for creating the virtual environment, only a link to the Python venv documentation. The conventional invocation on the Python side is python -m venv .venv followed by activation, but that command is not in the repository, so treat it as general Python knowledge rather than project documentation.
Where a link list stops being enough
The failure mode is link rot and drift. Every roadmap is a pointer to something hosted elsewhere: a book, a YouTube playlist, a university course, a paper. The repository cannot detect when a course is retired, a playlist is made private, or a paper's method has been superseded. Nothing in the material describes a link checker, a scheduled job or a CI step that validates the URLs.
The second limitation is that a reading list cannot tell you when to stop reading and start building. The README says the resources balance theory, intuition and real-world application, and that recommended items are marked with a star. That marking is an editorial judgement made by the maintainers, and the supplied material gives no criteria for it beyond the phrase high-impact content widely used in academia, research and industry.
The third is scope. The repository covers AI safety, alignment, governance and policy in one roadmap file. A learner who wants to work on alignment research and a learner who wants to be a computer vision engineer are sent down paths that share only the foundations. The career table acknowledges this by giving separate sequences, but the depth of each path is bounded by what a single Markdown file can hold.
What it is not competing with
The obvious alternative is a structured course with graded work, such as the fast.ai Practical Deep Learning for Coders course, which pairs video lessons with notebooks you run and modify. The difference in approach is the direction of the dependency: fast.ai owns its exercises and updates them together with the lessons, while this repository delegates all content to third parties and owns only the ordering. If a linked course changes its syllabus, the roadmap's ordering may no longer match it.
A second alternative is a single comprehensive textbook used as a spine, for example a standard machine learning text, with the repository's tables serving as supplementary reading. That inverts the repository's role: instead of being your plan, it becomes a bibliography. For someone who already has a spine, that is the more honest use of it.
The repository's advantage over both is breadth per unit of effort. One clone gives you paths into vision, NLP, LLMs, RAG, MLOps and safety without you assembling the list. The cost is that none of those paths is verified by execution.
Licence and maintenance
The repository is MIT licensed. For a collection of Markdown files that is permissive: you can copy the roadmaps into your own notes, fork them for a study group, or republish them with attribution. MIT covers the repository's own text. It does not cover the books, courses, papers and playlists the roadmaps link to, and the supplied material does not describe how those are licensed. Do not read the MIT file as permission to redistribute the linked material.
Maintenance is the open question. The last push recorded is 2026-06-15, and no releases have been retrieved, which is expected for a repository that ships no versioned artifact. There is a Contributions Welcome badge and a Contribution section in the table of contents, so the intended model is pull requests against the Markdown. That model works for adding links and fails quietly for removing dead ones, because nobody gets a notification when a third-party course disappears. If you fork this and depend on it, the upkeep is yours: re-check the roadmap files you actually follow, on a schedule you set, and fix the entries that no longer resolve.
Editorial conclusion
Adopt this if you want a single directory of Markdown roadmaps to work through in order, and you are willing to check each linked book, course or paper yourself before committing months to it. Do not adopt it if you need graded exercises, a runnable environment or a maintained dependency you can pin. Before you start, open roadmaps/ai-roadmap.md and Packages.md and confirm the Python 3.10+ setup table and the linked resources still resolve, because nothing in the repository enforces that.
Community notes