Build a neural network you can actually explain

Lots of people use neural networks without knowing what happens inside them. This tutorial fixes that—not with a giant framework, but with a tiny network you write in plain Python.

What you’ll build

A model that predicts whether a student passes based on study hours (and later, sleep hours too). Four rows of data. One neuron. Real training loop. No magic.

Study Hours	Passed
1	0
2	0
3	1
4	1

This is not production ML. It is a learning example so you can see every step.

What you’ll learn

• What a neuron does (input, weight, bias, output)
• How sigmoid turns a raw number into a probability
• What loss means and why training tries to shrink it
• How a simple loop updates weights—an honest preview of backpropagation
• How adding a second input changes the same math

Tutorial structure (7 pages, ~45 minutes)

1. The problem & the idea — dataset, what a neural network is
2. One neuron — forward pass by hand, first exercise
3. Sigmoid & decisions — probabilities, threshold, honesty about “thinking”
4. Loss & training — squared error, training loop, live code
5. Check your model — predictions, tuning exercises, short quiz
6. Two inputs — study + sleep, tabs for pure Python vs NumPy peek
7. Backprop, limits & final quiz — what we skipped, recap, knowledge check

Interactive pieces along the way

• Animated flows and diagrams for forward pass and training
• Drag-and-drop: order the training steps
• Live Python runners you can edit and run
• Quizzes on pages 5 and 7

Prerequisites

• Basic Python
• Simple math (multiply, subtract, square)
• No TensorFlow, PyTorch, or deep learning background required

Code repository

Runnable samples live in githubRepo/2026/05/20/tiny-neural-network-by-hand/. Same code as the tutorial—copy, run, break things on purpose.

---

---

Quick preview

A neural network here is just small math units chained together. Each unit does:

output = input × weight + bias

Training means: predict → measure error → nudge weights → repeat. That pattern scales all the way up to huge models. You’re learning the core loop first.

Ready? Hit Start Tutorial above.