Edge AI

sensor -> CRT -> classify

BME280 sensor on ESP32. Temperature, humidity, pressure decomposed into 7 CRT channels. Coupling determines kingdom. No training. No model. Real data from live boards.

Real Hardware

Three ESP32 boards on the CRT mesh. Two equipped with BME280 sensors (temperature, humidity, pressure). Data recorded from live hardware.

BoardElementKingdomSensor
FAR (COM3)sigma = 1bacteriaBME280
MIDDLE (COM4)D = 2protistaBME280
CLOSE (COM5)void = 0voidnone

Sensor Data Through CRT

Each reading is an integer. CRT splits it into 7 independent channels: 4 data (mod 8, 9, 25, 49) and 3 parity (mod 11, 13, 17). Green background = parity channel.

Readingmod 8mod 9mod 25mod 49mod 11mod 13mod 17
sigma 25.12 C0112134313
sigma 45.25 %57017413
sigma 1024.6 hPa642155212
D 23.92 C0717405012
D 47.82 %637298115
D 1024.4 hPa421933010

Kingdom Classification

coupling(n) = N / gcd(n, N). One GCD computation classifies any element into one of 8 kingdoms. No training data. No gradient descent. The ring structure IS the classifier.

Elementgcd(n, N)CouplingKingdom
sigma = 11214,414,200bacteria
D = 22107,207,100protista
void = 0214,414,2001void
No Training Required
Traditional edge classifiers need labeled data, gradient descent, and quantization to fit on microcontrollers. CRT classification uses algebraic structure: coupling(n) = N/gcd(n,N) maps any element to one of 8 kingdoms. No weights. No epochs. No overfitting. The algebra IS the model.

Classify Any Element

Enter any number. See its CRT decomposition, coupling, and kingdom.

Element 0
Coupling: 1
Kingdom: void (zero)
ChannelResidueType
mod 80data
mod 90data
mod 250data
mod 490data
mod 110parity
mod 130parity
mod 170parity

How It Works

BME280 Sensor
I2C, $2
Temperature + humidity + pressure at 0.01 C resolution. GPIO21/22. One reading every 3 seconds.
CRT Decompose
7 channels
Each reading modulo {8, 9, 25, 49, 11, 13, 17}. Four data channels carry the measurement. Three parity channels protect it.
Classify
O(1)
coupling(n) = N/gcd(n,N). One GCD computation. 8 coupling classes from 7 prime factors. No weights. No model.
ECC3 Verify
Rate 4/7
Triple-parity: reconstruct from 4 data channels, verify against 3 parity channels. 100% single-channel correction on hardware.
Total Size
536 bytes
Classifier WASM on ESP32. 240MHz, 520KB RAM, $4 board. No cloud. No GPU. No framework.

Source code · Public domain (CC0)

Report issue

.ax source compiled to WASM via self-hosting compiler. Zero HTML authored.