On-device audio classifier for theESP32-S3 that detects whether an electric pump is running. Captures stereo I2S audio, runs a quantised INT8 CNN on mel-spectrograms in ~31 ms per 1-second window, and reports ON/OFF state over TCP or MQTT. Includes Rust firmware, host-side receivers, PyTorch training pipeline, and a CT-clamp PCB

no_std Rust neural network inference engine for ESP32 microcontrollers. Compile-time model embedding, INT8 quantization, SIMD-optimized kernels, zero heap allocation

Native Rust exporter for converting Burn model graphs into ESP-DL .espdl files for ESP32-S3 INT8 inference