Software · Data and AI
Computer vision
Camera-based software that inspects, counts and detects, running on edge devices beside the camera or in the cloud.
Many visual checks are still done by tired eyes late in a shift. We build vision systems that inspect parts for defects, count items on shelves and conveyors, read labels and gauges, and check safety rules such as helmets in marked zones. Models are trained on images from your own site, and borderline cases are shown to a person rather than silently passed.
What we build
- Defect inspection on production lines for scratches, dents, misprints and missing parts
- Shelf and conveyor counting that reconciles physical stock with your inventory records
- Label, barcode, meter and number-plate reading where handheld scanners are impractical
- Safety monitoring for helmets, vests and restricted zones, with alerts to supervisors
- Edge deployments on NVIDIA Jetson or industrial PCs that keep working offline
- Labelling and review tools so your team can correct the model and add new defect types
DemoInteractive · sample data
Try it live.
A working sample with made-up data. It runs in your browser; nothing is sent anywhere.
- Line 2 · Batch B-4471
- Cap present
- Cap alignment
- Label position
- Batch code print
- Fill level
What it does well
Trained on your images
Models learn from your own products, lighting and camera angles, not generic stock datasets.
Edge or cloud
Inference runs beside the camera for speed and privacy, or centrally when sites share a model.
Review queue
Low-confidence detections are saved with the image so a person can confirm or overrule them.
Measured on real batches
Precision and miss rates are reported on held-out images from your line before go-live.
Privacy controls
Faces can be blurred on the device and footage kept only as long as your policy allows.
PLC and MES links
Reject signals go to the line controller and results are logged against batch numbers.
Questions
How many images do we need to train a model?
A few hundred labelled examples per defect type is often a reasonable start, but rare defects need deliberate collection. We begin with a short trial on your line, measure how the model performs, and tell you plainly which defects need more examples.
Will it replace our quality inspectors?
It works best taking the repetitive checks off their hands. The system inspects every item, flags the doubtful ones and records image evidence; inspectors handle flagged cases, new defect types and the final call on borderline batches.
Can it work with our existing cameras?
Sometimes. CCTV cameras are usually fine for counting and safety zones, but defect inspection often needs fixed industrial cameras and controlled lighting. We assess your setup with sample footage and recommend the least hardware that gives reliable results.