TinyML & Edge AI
Machine learning on tiny devices
Give your embedded projects intelligence. Learn how to train models and run them directly on microcontrollers — no cloud required.
Modules
A path from ML fundamentals to production-ready edge intelligence.
Beginner
ML Foundations for Engineers
What machine learning really is, when to use it instead of plain logic, and the maths you actually need.
- Supervised vs unsupervised
- Features & labels
- Training vs inference
- Overfitting
Beginner
Sensor Data & Feature Engineering
Collect, label and clean data from accelerometers, microphones and environmental sensors on your own boards.
- Data logging
- Sampling & windowing
- FFT & MFCC
- Normalisation
Intermediate
TinyML on Microcontrollers
Train in TensorFlow, then quantise and deploy to ESP32, Arduino Nano 33 BLE and STM32 with TFLite Micro.
- TFLite Micro
- Quantisation
- Model size vs RAM
- Latency tuning
Advanced
Edge Vision
Image classification and object detection on ESP32-CAM and Raspberry Pi with real-time constraints.
- Image pipelines
- Transfer learning
- ESP32-CAM
- FPS optimisation
Advanced
Audio & Keyword Spotting
Build always-on wake-word and sound classification devices that run on battery power.
- Keyword spotting
- Noise robustness
- Low-power always-on
- On-device buffers
Advanced
Predictive Maintenance
Detect anomalies in vibration and current signatures to predict machine failure before it happens.
- Anomaly detection
- Vibration analysis
- Thresholds & alerts
- Cloud dashboards
The edge ML workflow
01
Collect
Log real sensor data from your device — not a public dataset.
02
Train
Build and evaluate a small model in Python / TensorFlow.
03
Optimise
Quantise and prune until it fits in kilobytes of flash and RAM.
04
Deploy
Flash it to the MCU and run inference at the edge, offline.