Skip to content
EVENOR

Proof of concept

Assessment support for descriptive answers

Descriptive answer scripts take a long time to mark, and consistency drifts between the first script of the day and the hundredth. Teachers want the time back, but nobody is willing to hand a student's grade to a machine, and rightly so.

Layers of data being checked before a model is trusted

What we built

  • Reading of scanned answer scripts including handwritten answers, with the page kept alongside the text
  • A suggested band against the department's own rubric, with the sentences that justified it highlighted
  • Consistency checks that flag scripts marked very differently from similar answers
  • A marking screen where the teacher confirms or changes every mark before it counts
  • A record of what was suggested and what the teacher decided, so the two can be compared over a term

What it showed

The proof of concept produced a first pass that teachers could accept or correct quickly, and made drift visible across a batch. Every final mark was set by a person. The system never grades on its own.

What a live engagement needs

A first engagement needs the rubric as the department actually applies it, a sample of already marked scripts to calibrate against, and a written position for students on how AI is used in marking.

Services behind it

AI workflow automationData and AI foundations
Talk about your own case

Related proof of concept work

See the whole library