Build an AI camera and fridge demo
Start with the HardwareBlocks Camera Kit for a private snapshot dashboard. To turn an image into a draft shopping list, your own backend retrieves the image, calls a vision model and asks the user to review the result. HardwareBlocks supplies the camera-data interface; AI inference is a separate integration.
Build it step by step.
Design the software demo
Create a Camera simulation. It records example frame metadata, not a real photograph. Use a clearly labelled sample image in a custom application to demonstrate the intended vision flow.
Connect a physical camera
After provisioning, claiming and Wi-Fi setup, firmware uploads a JPEG to /api/v1/snapshot using its device token. Maximum upload is 256 KiB; only the latest image is retained.
Keep images private
Fetch /api/v1/devices/{id}/image with the owner’s workspace token from a trusted backend. There is no public image URL. Keep workspace and model-provider keys out of browser code and prompts.
Review the AI output
Ask your chosen vision service for a draft list. Let the user correct missing, obscured or misidentified items. Do not infer food freshness or consumption safety from a photo.
Test the fridge environment
Test lighting, reflections, mounting, condensation, temperature, radio coverage and power. Trigger captures on a schedule in your firmware. A door sensor is not included.
What runs where
Camera → Wi-Fi → private snapshot API → your backend → vision model → reviewed shopping list. Continuous video streaming, built-in recognition, automatic door events and an included inference subscription are not part of the kit.
The event contract.
Example payload for simulation; it is not a physical measurement. Use the current sensor schema for validation.
{
"type": "camera.snapshot",
"data": {
"image_url": null,
"width": 640,
"height": 480
}
}Your starting prompt.
Copy into your coding assistant. Keep all private keys out of the prompt.