Our approach

What we believe

Deterministic, verifiable methods decide what happens next; the language model decides how it is said. Never the reverse.

What makes AI output trustworthy, in any field, is that a named person reviews and approves it. The system proposes, the person approves, and every approval leaves a record.

In education specifically, mastery is measured per concept, not per course. That is where this belief is most tested today.

The expert at the centre

This is structural, not a marketing position: the person who knows the work approves what the system produces, and we build our tools to save their time, not to take their place.

In education the expert is the teacher, who reviews and signs the curriculum graph the system extracts from their lectures. Our target is that this takes no more than twenty-five minutes per lecture-hour.

What we do not claim

Publishing our limits is the most honest thing we can offer people who have heard inflated promises before.

• No guaranteed results, and no improvement in grades or figures without a published study to show it.
• We do not say our systems know everything.
• Nor that they are never wrong. We say how an error surfaces, and how the expert corrects it.
• We do not use AI where a simpler solution will do.

How we measure ourselves

We name our measures publicly, and do not publish their current values until they hold. In education we measure:

• The time a teacher needs to review and sign off one hour of lecture.
• The precision of concept extraction from lectures.
• The share of answers flagged as unsupported by any source.

With every organisation we work with

We agree the measure before we start building, and report the result as it is. Every release is tested against fixed datasets, and any drop in quality stops the release until it is fixed.

Arabic first

We write every text in Arabic first and then translate it, and design interfaces right-to-left from the start. A system that does not work well in Arabic does not serve this region, however good it is in another language.