field guide · curriculum 2026 · always open

523 living lessons

Aye Intelligence

Build AI from raw math before importing a framework. Follow one open path from vectors and backprop to agents, infrastructure, and working capstones.

Start Phase 0 Browse lessons

No paywall. No mystery. Just build the thing.

Your notebook

0 of 523 lessons

20 connected phases

See the whole journey

Course-aware tutor

Ask what the lesson means.

Get an explanation grounded in the curriculum.

Ask AI

Fork · remix · contribute

The entire course is open source.

Open repository

Learning streak

0days

Complete a lesson to mark today.

One repeatable loop

Theory becomes something you can run.

Every lesson moves from intuition to an artifact you can test, change, and keep.

1Read

2Derive

3Code

4Test

5Keep

The full curriculum

Pick a phase. Start building.

Every count comes directly from the living course.

Search all 523 lessons

warm up the workshop0/12

Outcome: Set up a repeatable AI engineering workspace.

vectors, calculus, probability0/22

Outcome: Use the math that powers optimization and neural networks.

regression to ensembles0/18

Outcome: Train, compare, and improve classical machine-learning models.

backprop by hand0/13

Outcome: Implement backpropagation and a small neural-network framework.

CNNs, ViTs, segmentation0/28

Outcome: Build an end-to-end computer-vision pipeline.

tokens to transformers0/29

Outcome: Build systems that understand, retrieve, and generate text.

spectrograms, ASR, TTS0/17

Outcome: Create speech, audio, and voice-assistant pipelines.

attention, KV cache0/16

Outcome: Implement a transformer from attention through inference.

diffusion, VAEs, GANs0/15

Outcome: Build and evaluate modern generative models.

policy & value methods0/12

Outcome: Train agents to learn policies from rewards.

build a tiny GPT0/24

Outcome: Assemble and train a small language model from scratch.

prompts, RAG, evals0/17

Outcome: Ship a grounded LLM application with evaluation and guardrails.

vision + language + audio0/25

Outcome: Build AI that reasons across text, images, video, and audio.

MCP, function calling0/31

Outcome: Create secure tools, protocols, and MCP integrations.

loops, memory, planning0/54

Outcome: Build an evaluated agent with planning, memory, and tools.

long-running agents0/22

Outcome: Operate long-running autonomous systems with safety controls.

cooperation & roles0/25

Outcome: Coordinate specialized agents toward a shared result.

serving, scaling, ops0/28

Outcome: Deploy, observe, scale, and optimize production AI systems.

the human side0/30

Outcome: Evaluate and reduce safety, alignment, and governance risks.

ship something real0/85

Outcome: Ship a portfolio-ready AI system end to end.

Certification paths

Turn the curriculum into a focused exam study plan.

Explore tracks

The workshop command

$ git clone https://github.com/rohitg00/ai-engineering-from-scratch
$ cd ai-engineering-from-scratch
$ open phases/00-setup-and-tooling

# Read. Derive. Code. Test. Keep.