Ahmed Hmeid Contact
All work 03 · AI writing tutor

Rook Learning

Good writing is rewriting. Writing practice for Years 4 to 11, where the AI marks and the child redrafts.

Open rooklearning.hmeid.uk

What it is

Rook sets children structured writing, marks each answer straight away against a rubric for their school year, and gives a reason for every mark. Then the child improves it and sees exactly what moved. The AI never writes a word of the answer: the prompts forbid it, and code checks the feedback and holds back anything that gives the wording away.

  • 80rubrics across Years 4 to 11
  • 640writing prompts in 8 genres
  • 3xcheaper marking after a measured redesign
  • 300+automated tests

Key features

  • Mark, then redraft

    Scores out of 10 for each check with a plain-English meaning, the one to three things to work on next, and "Improve this answer" to try again and see each score move.

  • A rubric for every year

    Written against the National Curriculum in England for each key stage, up to GCSE assessment objectives for Years 10 and 11.

  • Writing and reading

    Eight genres of writing prompt, plus reading lists of real school texts with chapter check-ins, judged only on the chapters read so far.

  • Bring your own work

    Students can bring in school work from any subject, pasting it or snapping a photo. The photo is transcribed and then discarded.

  • A bookcase of progress

    Finished books go onto an animated bookcase. Open one to page back through everything written about it.

  • Parents and tutors

    Child accounts without email, dashboards and charts, assigned work, custom exercises, and the ability to override a mark.

In action

The home page leads with the idea behind Rook: good writing is rewriting.
A Year 7 student writes a first draft describing a town street in a heatwave.
The AI marks each check out of 10 and quotes the child's own words as evidence.
After the redraft, every check shows how far it moved, here 6 to 9 and 4 to 9.
Each kind of writing becomes an exercise book on the student's bookcase.
Parents see which redrafts beat the first attempt and by how much.
A skills chart shows a parent where their child is strong and where to focus.
Tutors can override any AI score and leave a note the student will see.
On a phone, the redraft result celebrates the move up to Exceptional.
Feedback on a phone names the one thing to work on next.
The bookcase works on a phone too.
Under the hood

The parts that aren't obvious

  1. A marking lab, then a redesign

    An admin lab has the AI write 40 sample answers at planned quality levels, then marks them with competing designs. One grouped call beat one-call-per-criterion: $0.026 and 20 seconds per marking instead of $0.081 and 34 seconds, with equal or better accuracy.

  2. Hybrid marking

    British spelling, length and punctuation are scored in code. Claude judges the rest, then code re-marks anything it skipped and caps content scores when the answer misses the task.

  3. "Never writes for them", enforced in code

    A guard detects feedback that hands over wording. It is rewritten once, or withheld.

  4. No copyrighted book text

    About 190 hand-built fact sheets tagged by chapter replace the book text, so a check-in on chapter 3 is never judged against chapter 10.

  5. Safe by design

    Essays are fenced off from instructions to block prompt injection, a screen flags concerning content, and row-level security plus database triggers enforce who can change what.

Built with

App
  • Next.js 16
  • React 19
  • TypeScript
  • Tailwind CSS v4
  • shadcn/ui
  • Framer Motion
  • Recharts
AI
  • Claude Sonnet 5
  • Web search tool
  • Vision transcription
Data
  • Supabase
  • Postgres + RLS
  • pg_cron
Quality and hosting
  • Vitest
  • Docker
  • Coolify