Products
Solutions
Published
4 June 2025
Written by Emily Curryer
Engineers are used to working with microcontrollers. AI accelerators, not so much. But with Edge Impulse machine learning the game has changed – and it’s simple.
At Hardware Pioneers Max 2025, we caught up with Ivan from the Edge Impulse. The senior solutions engineer showed us how its cloud platform enables engineers to train ML models using sensor data, such as audio, video, or environmental, and deploy it straight to the edge. That includes everything from industrial PCs to wearables and even mobile phones.
Edge Impulse’s pièce de résistance is its hardware-agnostic approach. Whether you’re building a camera-based safety system or a gas detection sensor, the platform helps optimise your model to run on whatever silicon you’re using. And if your device compiles C++ code, you’re all set.
To begin, collect your data. Edge Impulse machine learning tools help you import datasets from a variety of devices. Then, build an “impulse”. This is a pipeline that processes data and runs it through a model on-device.
A standout demo involved detecting cars in real time using a YOLO Pro object detection model—custom-built by Edge Impulse—and running it completely locally on Qualcomm hardware. The platform passed cropped frames into a visual language model running on the same board, enabling natural language queries, “what colour is the car?” or “what’s the number plate?”
For the everyday engineer, you don’t need to wait. You can collect training data using your phone, build a model, and test it right on the device. Need to spot keywords in audio or detect anomalies in vibration data? Not a problem. All this without needing to write a single line of low-level code.
Whether you’re a startup prototyping wearable health tech or an engineer retrofitting smart sensors into factory lines, Edge Impulse machine learning makes AI accessible, flexible and fun. Give it a go—the future of embedded design is happening right now, at the edge.
Comments are closed.
Comments
No comments yet