Deep Learning From Scratch
TinyCNN
A working neural network written from nothing but raw maths — living proof that when I say I understand AI, I mean every line of it.
// The Problem Everyone Uses AI. Almost No One Can Explain It.
Today, anyone can copy ten lines of someone else's code and call themselves an AI builder. It feels like knowledge — right up until a recruiter, an interviewer, or your own curiosity asks one simple question: "but how does it actually learn?" And the honest answer is silence.
Most tutorials only show you which buttons to press. They never open the machine: why a network improves, why it suddenly gets worse, or what all the numbers inside are really doing. You end up driving a car you couldn't draw from memory — powerful, until something breaks or someone asks you to prove you know what's under the hood.
TinyCNN started from that exact discomfort. If I couldn't explain it, I didn't truly know it. So instead of leaning on ready-made tools, I set out to build the machine myself — from nothing but mathematics, one piece at a time.
// The Payoff What You Get When Nothing Is Hidden
The Black Box, Opened
You can watch a network learn with zero magic — every step happens in code you can actually sit down and read.
Maths Made Visible
Each idea behind modern AI is written out by hand and tested on its own, so nothing "just works" without a reason you can point to.
Proof of Real Fundamentals
Anyone can call a library. Building one's insides from scratch is the kind of evidence interviewers and mentors actually trust.
An Honest Roadmap
This project grows in public. What's finished is listed, what's next is stated plainly — convolutional pieces are underway, with more planned.
Built to Be Learned From
Every part comes with the thinking behind it, written clearly enough for students and curious minds to follow along.
Alive and Growing
New pieces are added regularly, complete with the failed attempts and quiet victories that come with building something real.
Already built and tested by hand:
// How It Works How a Network Actually Learns
Take a Guess
Information flows forward through layers that were written by hand, each one reshaping it until the network produces an answer.
Measure the Miss
That answer is compared against the truth, and the size of the mistake is boiled down into one clear score.
Blame, Then Adjust
The score travels backwards through the network, and every number inside learns exactly how much it contributed to the error — and which way to move.
Repeat Until Skilful
Run this loop over and over and tiny corrections stack up: wild guesses slowly turn into reliable predictions you can watch happen live.
// Who It's For Made for the Curious
Don't Just Use Neural Networks. Understand Them.
Open the live demo and watch a hand-built network learn in front of you — or read the code line by line and check for yourself that nothing is hidden.