Interactive neural network concepts for developers
No ML background required. Start by debugging a decision.
Your first neural network bug
The deploy is safe.
The neuron says BLOCK.
Every signal is green, but one connection learned the wrong direction. Flip a single weight to fix the decision.
CHALLENGEWhich connection is backwards?
live inference
bias = −1.601
tests_pass1
latency_stable1
security_clean1
rollback_readyσ(weighted sum)43%BLOCK
sum = -0.30What just happened?
You changed one number.
The machine changed its mind.
A neuron is surprisingly familiar: inputs come in, each gets multiplied by a weight, a bias shifts the default, and an activation turns the score into a decision.
Open the 5-minute neuron lab ↗input≈variableweight≈learned configbias≈default offsetactivation≈decision gateA visual learning pathThree concepts.
Three concepts.
One mental model.
Each page teaches one idea with one experiment. No setup, notation wall, or toy chatbot.
015 min lab
How does evidence become a decision?
How a neuron makes a decision
026 min lab
Why can’t one neuron solve every pattern?
Why neural networks need hidden layers
035 min lab
Why introduce a bend into the network?
Activation functions
Modern AI feels mysterious because we usually meet the output first.