Doodle AI

Welcome to Doodle AI! This is not a simple drawing app, nor a pre-programmed "smart" piece of software. It is an empty digital brain, much like a newborn: at first it knows nothing about the world, but it can learn.

Your job is to teach it to recognize your handwriting.

🚀 How to use it (quick start)

The AI starts out "dumb." Draw it a digit and it will guess at random. Don't worry — that's how it works! Here's how to teach it:

1. Draw: draw a digit (say, a 1) in the black square on the left with your mouse or finger.
2. Check: look at the guess on the right. (It will probably be wrong at first.)
3. Teach:in the button row below, click the digit you actually drew. Behind the scenes the AI quickly "re-reads" your drawing 40 times and adjusts its brain.
4. Repeat: clear the board, draw another digit, and click its button.

Practice: do this with a few digits, 3–4 times each. Then try drawing the first one again — now it will recognize it!

Tip: once you are happy with the result, click "💾 Save Memory." This downloads the trained brain, which you can load back any time later.

🧠 What happens behind the scenes? (How it works)

This program simulates an artificial neural network. To picture how it works, imagine a complex network of pipes with water flowing through it.

1. The input (the "eye")

When you draw on the grid, every pixel is an input signal. A white pixel "opens the tap" and lets the signal flow on; a black pixel keeps the tap closed.

2. The hidden layer (the "brain")

The dots and lines in the middle panel are the neurons. The secret is in the lines: these are the weights. A thick green line means "this connection matters, amplify it"; a thick red line means "this is forbidden, block it." At first these lines are thin and random — which is why it guesses badly in the beginning.

3. The output (the "mouth")

The signals flow through the network and arrive at the 10 outputs on the right (0 through 9). Whichever collects the strongest signal becomes the machine's guess.

🎓 Glossary — what does the "Teach" button do?

When you click the correct digit, a process called backpropagation starts — the "magic" of artificial intelligence. The machine compares its own guess with what you told it, sees how large the error is, then walks backwards through the network and nudges every line (weight) a little — strengthening the important connections and quieting the ones that fired by mistake.

This happens in a fraction of a second, but mathematically it is very complex. This is how "knowledge" takes shape in the thickness of the lines.

Why is this better than traditional programming?

In traditional programming the developer would have to write: "if there is a vertical line in the middle, it's a 1." But what if you write it slanted? Or with a serif? The program breaks. With this approach the machine learns your writing style instead. It doesn't memorize rules, it recognizes patterns — which makes it far more flexible.

Have fun teaching! Remember: there are no wrong answers, only a network that isn't trained enough yet! 😉