Neural AI Works.
ENG
ENGKOR

The AI Architect's Path Series

STAGE 1The Starting Point of AI

Build the mathematics behind AI, one idea at a time. This book puts you to your first test.No prior knowledge required. Just be ready to think it through.

  • Chapters 1–4 for free
  • Beginner
  • No setup
STAGE 1: The Starting Point of AIRead Chapters 1–4 for free ahead of the full release.
CHAPTERS 1–4 FREEEARLY ACCESS

Do you know the answers to these questions?

  • Why are GPU prices rising?
  • Why do we need partial derivatives?
  • Why does Micron’s stock rise when demand for AI GPUs surges?
  • What mathematical principle allows AlphaGo to predict its next move?
  • What conditions must an activation function satisfy?
  • Why can’t neural networks be nonlinear from the start?

Inside the book

Chapter 0.
To Aspiring Architects Standing at the Starting Point

AI is shaking up the world.

Everyone is talking about AI, and everyone wants to learn AI.

New models and new technologies emerge without pause, and lectures and books explaining them pour out day after day.

Yet when we delve into AI to truly understand it, we discover something a little strange.

“Contrary to what we often imagine, AI is not neatly organized into a single, unified discipline.”

At least, that is the case if we expect a fully formed discipline like physics or biology, where knowledge has been organized over a long period of time and what to learn, and in what order, has been thoroughly worked out.

In other words, it is hard to say that this field currently has a single, complete map that everyone can simply follow in the same way.

Perhaps this is only natural.

Modern AI is still changing rapidly.

Even as new knowledge is discovered, organized, and refined into a form that can be passed on to someone else, technology continues to move forward.

The roads are being built faster than the map can be drawn.

Yet if we look deeper, we discover that the opposite is also true.

6-STAGE AI BOOK SERIES

From a Linear Model to the Transformer, A 6-STAGE Journey Through the Architecture of AI

This book is organized into six stages, with each stage peeling back another layer of AI and gradually leading to modern AI.

Stage 1 begins with the linear model, one of the fundamental building blocks of deep learning. From there, the series progresses through MLPs, CNNs, RNNs, Transformers, and beyond.

No-Code Edition

A Path to Understanding Without Code

The No-Code Edition is for readers who want to follow the structure and flow of AI without writing code.

Through graphs, illustrations, quizzes, and interactive elements, you build an intuitive understanding of why deep learning models work the way they do.

Recommended for: non-developers · product managers ·
educators · AI beginners

Code Edition

A Path to Understanding Through Code

The Code Edition is for readers who want to explore AI concepts through code and live results.

Run code directly in your browser to trace the model’s computations and training process.

FREE EARLY ACCESS

STAGE 1The Starting Point of AI

This is where those who learn AI and those who build AI part ways. This book puts you through your first test.

No prior knowledge required. Just be ready to think it through.No prior knowledge required. Just be ready to think it through.

  • For beginners exploring a path in AI
  • For developers rebuilding the foundations

Free Early Access for the first 100 readersRead Chapters 1–4 for free ahead of the full release.

Up to 70% off for completing the surveyReceive a discount on the full STAGE 1 release.

Read Chapters 1–4 for FreeEARLY ACCESS

How learning works with AI Book

Do more than read. Explore, check, and ask.

y = wx + b
READ

Build context through story

The problem appears before the formula, so every idea arrives after you know why it is needed.

6 STAGES, one connected journey

Each stage carries what you learned into the architecture that follows.

STAGE 1The Starting Point of AI

  • Build the mathematics behind AI, one idea at a time.
  • No prior knowledge required.
  • Just be ready to think it through.
Read FreeEARLY ACCESS
STAGE 2
MLP

Designing a multilayer perceptron

  • Build hidden layers
  • Learn through backpropagation
  • Design a deep network
Planned
STAGE 3
CNN

The geometry of convolution

What Makes This Book Different

The math behind deep learning, built for beginners.

You do not begin with derivatives, gradients, or loss functions as abstract formulas. You begin with simple problems you can reason through, then build step by step toward the mathematics that powers deep learning.

You understand the problem before you learn the formula. A new mathematical idea appears only when the tools you already have are no longer enough. By the time derivatives or gradient descent arrive, you already understand why they are needed.

Each idea grows naturally from the one before it. Rates of change lead to derivatives. Derivatives lead to gradients. Gradients lead to gradient descent. Instead of learning isolated formulas, you see how the pieces connect.

One continuous story keeps the math grounded. The same characters, decisions, and data continue across chapters. As the problems become more sophisticated, the mathematics grows with them—without asking you to start over each time.

Who this book is for

  1. People deciding whether AI study is right for them
  2. Complete beginners who want to start from first principles
  3. Undergraduates who want to connect formulas, code, and model behavior
  4. Career changers building a reliable foundation before switching fields
  5. Developers who want to understand why models learn, not just how to run them
  1. People deciding whether AI study is right for them
  2. Complete beginners who want to start from first principles
  3. Undergraduates who want to connect formulas, code, and model behavior
  4. Career changers building a reliable foundation before switching fields
  5. Developers who want to understand why models learn, not just how to run them

About the author

R

Richard One

Richard One is a PhD student with a deep interest in how AI is built from its most fundamental principles.He focuses on making those ideas accessible to people without a technical background.That interest led him to create this book.

The AI Architect's Path

Design deep learning from first principles, not outcomes

Explore the full series

Frequently asked questions

What is Early Access?

It is a free preview of Chapters 1–4 before the full release. Your feedback helps us refine the explanations, learning flow, and scope for the full release.

What can I read right now?

Chapters 1–4 are available now. Chapters 5–7 are planned for the full release and will continue on this same STAGE 1 page.

Do I need coding experience?

No. The No-Code Edition covers every core idea, and you can choose the Code Edition when you want to inspect the calculations.

Do I need to be good at math?

AI Book does not avoid equations, but it always builds intuition first with visuals and analogies. Concepts such as averages and slopes are explained again in context, so you do not need to know all the math before you begin.

How is this different from an e‑book or PDF?

You can adjust simulations inside the text, check your understanding with chapter quizzes, and ask questions about what you are reading. It is designed to be used, not just viewed.

Read Chapters 1–4 for free in Early Access

Read Chapters 1–4 before the full release and share your feedback.Free Early Access is available to the first 100 readers, and survey participants can receive up to 70% off the full STAGE 1 release.Read Chapters 1–4 before the full releaseand share your feedback.Free Early Access is availableto the first 100 readers,and survey participants can receiveup to 70% off the full STAGE 1 release.

FREE EARLY ACCESS

Get AI Book updates

Be the first to know when a new stage becomes available.

Notify me
Neural AI Works.
Privacy Policy© 2026 Neural AI Works, Inc. All rights reserved.

“Many of the principles at the foundation of today’s AI are by no means new.”

Linear algebra, calculus, probability and statistics, geometry and vectors.

Countless mathematical principles that power AI today existed long before AI itself emerged.

Humanity has long been developing mathematics to understand how the world works.

We have discovered, one by one, ways to explain motion, measure change, express spatial relationships, and handle uncertainty with numbers.

Countless mathematical discoveries—whose uses were sometimes unknown at the time—accumulated across generations, and new discoveries were built on top of them.

And the mathematics that had lain dormant for so long met new algorithms, vast amounts of data, and computing power far beyond what had come before.

“And it awoke.”

This book takes us back to that starting point.

We will not begin by memorizing the structure of a finished AI system laid out before us.

Nor will we begin by looking at formulas someone else has already devised and learning how to use them.

Instead, we will try the reverse.

“We assume that nothing has been built yet.”

How can we express a judgment in numbers?

How can we calculate the error in that judgment?

How can we find out in which direction a better answer lies?

We will try the simplest method, discover its limits, and devise a new one.

“You need no prior knowledge of mathematics. You need no advanced programming experience. You need only be ready to think things through to the end.”

As long as you do not stop, this book will show you the way to the very end.

September 2026 Richard One

Chapter 1. The Decision-Making Formula

1-1. A Formula for Travel Satisfaction

“Have you ever seen anything like this?”

Ari held out her phone.

A breathtaking sight unlike anything I had ever seen.

From the screen, a view so magnificent it could bring tears to your eyes was about to spill out.

“Looks great. But isn’t it pretty expensive to get there?”

“Sure, but if we go as a group, we can split things like the accommodation.”

Ari thought for a moment. A trip is not satisfying just because the destination is a good one. Satisfaction also depends on who comes along, how much it costs, and how long you stay. …

… For Ari, for example, having more Friends joining is a good thing, while a higher Estimated cost is a burden. So how can these two factors be expressed in a single expression?

In subtraction, the larger the first number and the smaller the second number, the larger the result.

For example,

10 − 2 = 8

10 − 7 = 3

When the first numbers are the same, the smaller the second number, the larger the result. And when the second numbers are the same, the larger the first number, the larger the result.

Using this property of subtraction, Ari set up a very simple formula: whatever raises her satisfaction goes first — call it the good factor — and whatever lowers it goes second, the burdensome factor.

Expected satisfaction = Good factor − Burdensome factor

Written this way, the larger the good factor and the smaller the burdensome factor, the higher the Expected satisfaction the expression produces.

For Ari’s trip, the good factor is Friends joining and the burdensome factor is Estimated cost. So the formula becomes:

Expected satisfaction = Friends joining − Estimated cost

Menu
Contents

Chapter 0.To Aspiring Architects Standing at the Starting Point

    Chapter 1.The Decision-Making Formula

    • 1-1.A Formula for Travel Satisfaction
    • 1-2.Applying Influence

    Chapter 2.The Error-calculation Formula

    • 2-1.Adjusting the Influence
    • 2-2.The Sum of the Errors
    • 2-3.The Sum of Squared Errors
    • 2-4.The Graph of the Sum of Squared Errors
    • 2-5.Distinguishing the Decision-Making Formula from the Error-Calculation Formula
    • 2-6.The Coordinates of the Vertex of the SSE Graph
    • 2-7.The Mean Squared Error
    • 2-8.Flexibility in Designing an Error-Calculation Formula

    Chapter 3.The Output-Adjustment Function

    • 3-1.Data That Does Not Form a Straight Line
    • 3-2.The Sigmoid Function
    • 3-3.Transforming the Sigmoid Function
    • 3-4.Composing the Decision-Making Formula with the Output-Adjustment Function
    • 3-5.The MSE of a Decision-Making Formula Composed with an Output-Adjustment Function

    Chapter 4.Measuring Change

    • 4-1.Calculating the Slope
    • 4-2.The Average Rate of Change and Its Sign
    • 4-3.The Slope Between Two Very Close Points
    • 4-4.How to Find the Slope at a Point

    Chapter 5.Finding a Low Point

      Chapter 6.Finding a Low Point with Multiple Variables

        Chapter 7.Toward "The Anatomy of AI"

          Recommended for: developers · data analysts ·
          Python learners           · developers new to AI

          w₁bMSE
          INTERACT

          Learn by doing

          Adjust simulations of learning rate, gradient descent, and other core concepts. Watch the loss curve change in real time and build intuition through direct experience.

          Q. How do you find the gradient of MSE(w₁, b) in the b direction?
          Fix w₁ → cross-section → differentiate with respect to b
          Fix b → w₁ cross-section → differentiate with respect to w₁
          QUIZ

          Learn by checking

          Use quizzes and term tooltips in every chapter to check what you understand, not just what feels familiar.

          How do the weight and bias learn together?
          Find the gradient for w₁ and b with partial derivatives, then update both values in the opposite direction of their gradients.
          AI

          Learn by asking

          Ask AI whenever you get stuck on a sentence. You can stay in the chapter, and every answer is grounded in the text you are reading.

          Start with the parts
        • Read spatial features
        • Design a visual network
          • Planned
          STAGE 4
          RNN

          The flow of recurrence

          • Work with ordered data
          • Remember what came before
          • Understand time series
          Planned
          STAGE 5
          LSTM

          Long-term dependencies

          • Remember and forget well
          • Choose with gates
          • Handle long sequences
          Planned
          STAGE 6
          Transformer

          Attention and the Transformer

          • Decide where to look
          • Connect key context
          • Process in parallel
          Planned
          Which device is best for learning?

          You can read on mobile, but a desktop provides the most comfortable experience for viewing interactive graphs alongside the full text.

          In what order will the stages become available?

          The series progresses from SLP to MLP, CNN, RNN, LSTM, and Transformer. Following that order helps you understand how model architectures have evolved.