Computer vision

With a task where someone looks at images or objects today to make a call: inspecting, counting, reading. On a first 30-minute call, we tell you honestly whether computer vision is the right answer, or whether a sensor or a simple rule would do.

Not necessarily. A smartphone or the cameras you already have are often enough to start. We check image quality during the pilot and only recommend hardware if the results call for it.

Not always. To read text or recognise common objects, existing models often do the job. For a defect specific to your products, we train a model on your own images, including examples of faulty parts.

It will happen, as it does to any person doing inspections. We measure the error rate before launch, set a confidence threshold below which a person checks, and keep tracking results over time.

Mostly the image work: how many images are needed, and whether they have to be labelled by hand, which means marking what each one shows. Then the number of sites, the hardware and the tools to connect. A narrow use case with photos you already have moves much faster than equipping a production line.

We favour hosting in Europe, or processing on site when possible. As soon as a person or a number plate is in the picture, we keep only what is needed, can blur faces and plates, and agree the rules of use with you under GDPR and the EU AI Act.