Services · AI and data
Computer vision
I train models to find defects, objects or changes in photos and video. The result shows exactly where the problem is, and uncertain cases go to a person for review.
ResultThe defect is marked in the image; uncertain cases go to a person.
Who it is for
- Products or goods are checked by eye, and the result depends on the person.
- Cameras record, but nobody watches the footage.
- There are thousands of photos that need sorting or checking.
What I can do
- Surface defect and reject detection
- Detecting and counting objects in video
- Sorting and checking photos automatically
- Marking a defect’s location at pixel level
- An explanation of what the model looked at
- Running in the browser, on a server or on site
Input → output
| Input | Output |
|---|---|
| A photo of a product on the line | “OK” or the defect’s location in the image with a confidence score |
| A warehouse camera stream | Counted objects and an alert on anomalies |
| 5,000 photos in one folder | Sorted into categories, doubtful ones flagged for review |
A typical format, not a client project.
Proof
- Founder’s independent projectSteelVision
A segmentation model (UNet++, EfficientNet) finds steel-surface defects and marks them at pixel level. Open the demo
- Founder’s independent projectSmartCam
YOLOv8 detects objects in a live video stream and shows the result right in the browser. Open the demo
- Founder’s independent projectOncoView AI
A Grad-CAM explanation shows, for every prediction, what the model relied on in the image (a public demo, not a diagnosis). Open the demo
How the work goes
- Understand the task. A free conversation about your workflow, programs and what gets in the way.
- Agree the proposal. I agree the outcome, scope, price and timeline in writing before starting.
- Build the solution. I show you interim versions and gather your feedback so the solution fits your workflow.
- Test with your data. Together with you I try typical work scenarios and fix any issues found before rollout.
- Hand over the system. You receive the solution, guide and access. I show you how to use it; ongoing support is agreed separately.
Price
- One automationfrom 490 €
- Projectfrom 2,000 €
- Maintenance and supportfrom 120 €/ per month
The first conversation is free. The exact figure depends on the number of sources, document types and special cases – you get it in writing before work starts. All prices →
Questions
How many images are needed?
To start, often a few hundred labelled examples per defect type. I review your images and tell you before work starts whether that is enough.
Do we need new cameras?
Not always. We first test with existing cameras or phone photos; if image quality is not enough, I tell you what to change.
What if the model makes a mistake?
Every result has a confidence score. Low-confidence cases go to a person, and those corrections help train the model further.
Let’s discuss your task
Write a few sentences about what is still done by hand. I reply personally.