Skip to main content
All example use casesIllustrative example

Arabic reading practice with teacher review

A learning centre wants an LLM to draft Arabic reading exercises, with a teacher checking everything before a child sees it.

The challenge

A learning centre for out-of-school children wants Arabic reading-practice exercises, generated and adapted to each child's level by a large language model. Teachers must review every exercise before it reaches a child.

Kind of support

  • Prompt design
  • Evaluation of Arabic language models
  • Child-safety guardrails
  • Teacher review workflow

How the mentorship could run

  1. Weeks 1 and 2: agree the reading levels, the types of exercise and the topics that are not allowed.
  2. Weeks 3 to 5: compare two or three language models on Arabic exercises, scored by teachers against a simple checklist.
  3. Weeks 6 to 8: write the prompts and guardrails for the chosen model, and build a review screen where teachers approve, edit or reject each exercise.
  4. Weeks 9 and 10: run the workflow with a small group of teachers, track how many exercises they change, and adjust the prompts.

What to have ready before you apply

  • Examples of good exercises at each reading level
  • The curriculum or reading framework you follow
  • How many teachers can review exercises, and how much time they have
  • Your safeguarding policy for children

Risks to plan for

  • Language models make mistakes in Arabic grammar and vowel marks. Teachers must check every exercise, and the review must be easy enough that they do it.
  • Content for children needs strict limits on topics. Write the list of excluded topics before you write any prompt.
  • Costs can grow with use. Estimate the cost per exercise early and set a monthly limit.

What a good result looks like

Teachers get a steady supply of exercises at the right level, spend their time reviewing rather than writing, and no exercise reaches a child without a teacher's approval.

More example use cases

  • Country: Bangladesh (Cox's Bazar)

    Sector: WASH

    Water-point dashboards with breakdown alerts

    A WASH NGO wants automated dashboards and alerts when a tap stand or latrine block stops working across 30 camps.

    Time: 1 to 2 hours per week, 8 weeksIllustrative example
  • Country: Colombia

    Sector: Cash assistance

    Finding duplicates across three cash programmes

    An INGO wants to find duplicate and likely wrong records across three programme lists, without building a surveillance tool.

    Time: 2 hours per week, 6 weeksIllustrative example
  • Country: Ukraine

    Sector: Shelter

    Building damage from drone and satellite imagery

    A shelter NGO wants to classify building damage from imagery so field teams visit the worst-hit areas first.

    Time: 1 to 3 hours per week, 12 weeksIllustrative example
  • Country: Nepal

    Sector: Information management

    A safe assistant over SOPs and guidance notes

    An information management officer wants to build a retrieval assistant over their organisation's SOPs, safely and on a small budget.

    Time: 1 hour per week, 8 weeksIllustrative example

Know more

Want to know more before you apply? Write to contact@theaiexpertgroup.com and a member of the Mentor Council will reply.