Ai Engineering From Scratch
Learn it. Build it. Ship it for others.
Last verified:
Visit Ai Engineering From Scratch
What is Ai Engineering From Scratch?
AI Engineering from Scratch is a free, open-source curriculum that teaches AI engineering by making you build core algorithms and systems from raw math before relying on frameworks. It is organized into 20 phases and 503 lessons, with material spanning linear algebra, machine learning, deep learning, transformers, LLMs, RAG, agents, multimodal systems, infrastructure, safety, and capstone projects.
The site emphasizes a hands-on workflow: read the problem, derive the math, write the code, run the test, and keep the artifact. Lessons are designed to ship runnable code and practical outputs, so the curriculum is not just theory but a build-and-verify path that aims to make you understand what tools like PyTorch are doing under the hood.
It supports four languages: Python, TypeScript, Rust, and Julia, depending on what best fits the lesson. The catalog also lets you search, filter, and sort lessons across all phases, which makes it easier to navigate a very large curriculum and jump to a specific topic or stage.
This tool is aimed at self-directed learners, aspiring AI engineers, software engineers, and advanced practitioners who want deeper fundamentals rather than shortcut tutorials. It is especially suitable for people who want to learn by building on their own machine, without a paywall, signup, or hand-holding.
Ai Engineering From Scratch pricing
Pricing model: Freemium
The website presents the curriculum as free and open source under MIT. It says there is no paywall and no signup, and the entire curriculum is available on GitHub to clone or fork and learn at your own pace. The site does not show any paid plans or tiered pricing, and it describes the material as runnable locally on your own laptop.
Ai Engineering From Scratch pros
- 503 lessons across 20 phases
- Covers fundamentals through production systems
- Teaches raw-math-first implementation
- Includes runnable code in each lesson
- Supports Python, TypeScript, Rust, and Julia
- Free and open source
- No paywall
- No signup required
- Runs on your own machine
- Searchable lesson catalog
- Filterable and sortable lessons
- Structured, phase-based progression
- Build-test-ship lesson loop
- Covers agents and autonomous swarms
- Includes capstone projects
- Uses practical artifacts per lesson
- Covers production infrastructure topics
- Includes safety and evaluation material
Ai Engineering From Scratch cons
- Very large time commitment
- Can be overwhelming for beginners
- No five-minute video format
- Little hand-holding
- Requires comfort with self-study
- Needs coding setup on your own laptop
- Some lessons assume mathematical maturity
- Not optimized for casual browsing
- Breadth may outpace depth for some topics
- No paid support or tutoring shown on the site
- Multi-language support may add complexity
- Advanced phases may be too ambitious early on
- Not a quick-start course
- Progress tracking appears minimal
- Best experience likely needs GitHub familiarity
Frequently asked questions about Ai Engineering From Scratch
What is AI Engineering from Scratch?
It is a free, open-source AI engineering curriculum designed to teach the field by building core algorithms and systems from the ground up. The site describes it as 20 phases and 503 lessons covering everything from math foundations to autonomous agent systems.
How is the curriculum structured?
The curriculum is organized into 20 phases, and each phase contains multiple lessons. The site says each lesson follows the same loop: read the problem, derive the math, write the code, run the test, and keep the artifact.
What topics does it cover?
It covers a wide range of AI engineering topics, including linear algebra, machine learning, neural networks, vision, NLP, audio, transformers, generative models, reinforcement learning, LLMs, RAG, tool use, agents, multimodal systems, production infrastructure, evaluation, and safety.
Which programming languages are used?
The curriculum uses four languages: Python, TypeScript, Rust, and Julia. The site says each lesson uses whichever language best fits the concept, and every lesson has runnable code in one of those languages.
Is it beginner-friendly?
It is beginner-accessible only if you are ready for a demanding self-study path. The site explicitly says it does not use short videos or hand-holding, and it is built for people who want to work through problems deeply and independently.
Does it require a subscription or account?
No. The website says it is free, open source, with no paywall and no signup required.
Can I use it offline or locally?
Yes. The site says it is built to run on your own laptop or machine, and it provides a GitHub clone command so you can fork or download the curriculum and work through it locally.
What makes it different from typical AI tutorials?
It is much more hands-on and foundational than a typical tutorial. Instead of just showing how to use APIs or frameworks, it focuses on deriving the math, implementing the algorithm, and understanding what frameworks are doing internally.
Does it include practical projects?
Yes. The curriculum includes lessons that produce reusable artifacts and culminates in capstone projects across areas like coding agents, RAG, multimodal systems, research agents, and distributed training.
How can I navigate such a large curriculum?
The site includes a lesson catalog for searching, filtering, and sorting all 503 lessons across the 20 phases. That makes it easier to jump to a topic, inspect what is complete, and choose a starting point that matches your current level.