AI Engineering from Scratch — Hands-On AI Systems Course

Introduction

AI Engineering from Scratch is a structured, open-source curriculum designed to teach developers how to build production AI systems from the ground up. Rather than relying on high-level abstractions, it guides you through implementing core components yourself so you understand what happens under the hood.

What AI Engineering from Scratch Does

• Provides 20+ hands-on modules covering transformers, agents, RAG, computer vision, and reinforcement learning
• Teaches building AI pipelines from raw math through to deployment
• Includes runnable Python notebooks for every concept
• Covers MCP integration, agent orchestration, and tool use
• Offers exercises that progress from basics to shipping production systems

Architecture Overview

The repository is organized as a series of standalone modules, each in its own directory with a README, Jupyter notebooks, and supporting code. Modules are designed to be followed sequentially but can also be used as standalone references. Dependencies are minimal and Python-based.

Self-Hosting & Configuration

• Clone the repo and install Python 3.10+ with pip
• Each module has its own requirements file for isolated dependencies
• GPU optional for most modules; recommended for vision and training sections
• Works on Linux, macOS, and Windows with WSL
• No external API keys required for core modules

Key Features

• Covers the full AI stack from linear algebra to multi-agent systems
• Emphasizes building from scratch rather than calling libraries blindly
• Includes real-world deployment patterns (Docker, cloud, edge)
• Community-maintained with regular updates for new techniques
• Rust and TypeScript supplementary tracks for systems-level AI

Comparison with Similar Tools

• Fast.ai — top-down approach vs this bottom-up from-scratch methodology
• Coursera/Udacity — paid and time-locked vs open and self-paced
• Andrej Karpathy tutorials — focused on neural nets vs full-stack AI engineering
• LangChain docs — framework-specific vs framework-agnostic principles
• Papers with Code — research-oriented vs engineering-oriented

FAQ

Q: Do I need a GPU to follow the course? A: Most modules run on CPU. GPU is recommended only for the training and vision modules.

Q: What prior knowledge is assumed? A: Basic Python programming and high-school math. Each module builds on the previous one.

Q: Can I use this for a university course? A: Yes, it is Apache-2.0 licensed and designed for educational use.

Q: How often is the content updated? A: The community contributes regularly, with major updates roughly monthly.

Sources

• https://github.com/rohitg00/ai-engineering-from-scratch
• https://github.com/rohitg00/ai-engineering-from-scratch#readme