How to Install AI Engineering from Scratch: Local, Skill Bundle, and Web Options

You can install the AI Engineering from Scratch curriculum by cloning the GitHub repository for local execution, installing it as a skill bundle via npx skills add rohitg00/ai-engineering-from-scratch for agent-driven learning, or accessing it directly at aiengineeringfromscratch.com.

The AI Engineering from Scratch curriculum by Rohit Ghumare is a modular, open-source educational repository structured into 20 phases covering foundational math to advanced agentic systems. Depending on your workflow, you can execute Python lessons locally, register the curriculum as portable SKILL.md agents within hosts like Codex or Claude Code, or stream the content through the static site generator.

Three Ways to Install the Curriculum

The repository supports three interchangeable consumption models. Choose the method that aligns with your tooling preferences:

• Clone & run locally: Download the full source tree to execute lesson scripts, inspect implementations, and modify code
• Install learning skills: Register agent-driven tutors that orchestrate the curriculum from within compatible AI hosts
• Read on the web: Access the lesson content through the browser without local file management

Method 1: Clone and Run Locally

Cloning the repository gives you complete control over the execution environment and allows you to run lesson scripts directly. This method requires Git, Node.js ≥20, npx, and Python 3.

First, verify your prerequisites as documented in the repository's README.md:

node --version
npx --version
python3 --version

Then clone the repository and navigate into the directory:

git clone https://github.com/rohitg00/ai-engineering-from-scratch.git
cd ai-engineering-from-scratch

The curriculum content resides in the phases/ directory, which contains 20 top-level phases. Each phase includes lessons following a uniform layout with code/, docs/en.md, and outputs/ subdirectories. To run a specific lesson:

python3 phases/01-math-foundations/01-linear-algebra-intuition/code/vectors.py

Method 2: Install as Learning Skills (Agent-Driven)

For users working within AI agent hosts like Codex, Claude Code, or compatible platforms, the curriculum can be installed as a portable skill bundle. This registers SKILL.md agents such as start-learning and learn-mcp that drive the interactive curriculum.

Execute the installer script via npx (implemented in scripts/install_skills.py):

npx skills add rohitg00/ai-engineering-from-scratch

After installation, invoke the interactive tutor using your host's specific syntax. For example, in Codex:

start-learning

You can also jump directly to specific lessons:

learn 01   # Starts the first lesson of the selected phase

The skill bundle references learning path manifests located in learning-paths/model-context-protocol.json, which defines the Model Context Protocol (MCP) routes and agent skills used by the curriculum.

Method 3: Read on the Web

If you prefer not to manage local files or dependencies, stream the curriculum directly at aiengineeringfromscratch.com. The static site is generated by site/build.js and deployed automatically via CI, offering the same lesson content without installation.

Repository Architecture and Key Files

Understanding the repository structure helps you navigate the curriculum effectively after installation:

• phases/ — Core curriculum containing 20 phases with standardized lesson directories (code/, docs/en.md, outputs/)
• scripts/install_skills.py — Python implementation of the skill bundle installer executed by the npx skills add command
• scripts/audit_lessons.py — CI-run validator that enforces lesson-structure rules and consistency
• learning-paths/model-context-protocol.json — Manifest file defining MCP and Agent Skills routes for the learning paths
• site/build.js — Static site generator that builds the interactive lesson viewer and documentation site
• README.md — High-level overview containing the phase-to-lesson mapping and quick-start commands

Verifying Your Installation

To confirm your environment is correctly configured for local execution:

# 1. Clone the repository

git clone https://github.com/rohitg00/ai-engineering-from-scratch.git
cd ai-engineering-from-scratch

# 2. Verify required tools

node --version
npx --version
python3 --version

# 3. Install the curriculum as agent skills (optional)

npx skills add rohitg00/ai-engineering-from-scratch

# 4. Test a specific lesson

python3 phases/00-setup-and-tooling/01-dev-environment/code/verify.py --route beginner

Summary

• AI Engineering from Scratch offers three installation methods: local Git clone, portable skill bundle via npx skills add, or web browser access
• Local execution requires Node ≥20, npx, and Python 3, with lesson code stored in phases/ subdirectories
• The skill bundle installer at scripts/install_skills.py registers agents like start-learning for host-driven curriculum navigation
• Learning paths and MCP routes are defined in learning-paths/model-context-protocol.json
• The static site at aiengineeringfromscratch.com is built from site/build.js and mirrors the local content

Frequently Asked Questions

What are the prerequisites to install AI Engineering from Scratch?

You need Node.js version 20 or higher, npx (included with Node), and Python 3. These tools enable both the skill bundle installation process and the execution of lesson scripts. Verify versions with node --version, npx --version, and python3 --version before proceeding.

What is the difference between cloning the repo and installing the skill bundle?

Cloning the repository downloads the full source code to your machine, allowing you to run Python scripts directly and modify lesson files. Installing the skill bundle via npx skills add registers SKILL.md agents within your AI host (like Codex) that orchestrate the curriculum interactively without requiring you to manually navigate file paths.

How do I run a specific lesson after installation?

For local execution, run Python files directly from the phases/ directory (e.g., python3 phases/01-math-foundations/01-linear-algebra-intuition/code/vectors.py). If using the skill bundle, use the host-specific command such as learn 01 to start the first lesson of your selected phase.

Where are the lesson files located in the repository?

All lessons are organized within the phases/ directory, which contains 20 top-level phase folders. Each lesson follows a consistent structure with a code/ subdirectory for executable scripts, docs/en.md for documentation, and outputs/ for expected results.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}