Much of an organisation's useful information does not live in a database: it sits on a photo, a scan, an exam paper, a label, a camera frame. Computer vision is how you make it usable.
What we do
- Document reading: structured extraction from invoices, forms and delivery notes, including ones photographed at an angle on a phone
- Handwriting: reading and interpreting handwritten text and reasoning
- Visual inspection: defect detection, compliance checks, counting
- Multimodal analysis: combining image and text to answer a precise question about what a photo shows
The case we know from the inside
Tenjah, our AI teacher for the Tunisian baccalaureate, grades a photographed handwritten paper in about thirty seconds. It is a hard problem: varied handwriting, crossings-out, mathematical notation, French and Arabic on the same page, photos taken under a desk lamp. That ground taught us where computer vision is reliable and where it needs a person behind it.
Our approach
We start by building an image set that reflects your real conditions — not catalogue shots. We measure results against a simple baseline before adding complexity. And we design the system to report its own uncertainty: an unreadable image should produce "needs review", not an invented answer.
What you get
A documented pipeline, metrics measured on your own data, and clear rules on what the system decides by itself and what it escalates to a person.
