How it works

From exam paper to learner — in five steps. Here is what each one involves, and roughly how long it takes.

1. Upload

A teacher or invigilator signs in to the Exam Reader client on any school PC and uploads the exam PDF. The server takes it from there.

The paper is split into readable units rather than treated as a wall of text. The segmenter identifies:

• section headers and question headers
• questions and their sub-questions
• multiple-choice options
• mark allocations, attached to the question they belong to
• the language of every segment — English or Afrikaans
• admin blocks, page footers and dot leaders, which are filtered out so they are never read aloud

That structure is what lets a learner move to “Question 4.2” instead of scrubbing back through an audio file.

2. Review

The teacher reviews the extracted content. The system is deliberately cautious. Anything it is not confident about is marked for human review rather than guessed at, and a paper cannot be assigned until those items are dealt with. Typically flagged:

• Visuals — diagrams, maps, photographs and graphs. The server runs OCR over the page image to recover any text inside the picture, then asks the teacher to write the description the learner will hear.
• Data tables — a table read left-to-right is meaningless, so the teacher decides the reading order and phrasing.
• Matching exercises — Column A / Kolom B layouts, where the pairing has to be spoken in a way that makes sense.
• Low-confidence text — unusual layouts, poor scans or mixed columns.

The teacher edits the spoken text directly, so an abbreviation, a formula or a currency amount can be phrased exactly as it should sound.

3. Approve

Nothing reaches a learner until a teacher has approved the questions and any flagged content. Once approved, the same wording is read to every learner, every time — without paraphrasing or interpreting the question.

Our benchmark for a workflow teachers will actually keep using: under 30 minutes for a Grade 4–9 paper and under an hour for Grade 10–12. If a paper is taking longer than that, we want to hear about it.

4. Assign

Once the paper is approved, the server renders audio for every spoken segment and caches it. This happens once, before the exam — not while thirty learners are waiting.

The teacher then assigns the exam to the learners approved for a reader in that sitting. Learners are imported from an Excel workbook or CSV file, pasted in from a table, or added by hand, with duplicate detection and required-field validation along the way.

5. Listen

The learner opens the same client app, chooses Student, enters their candidate number and confirms the paper. The app then locks into full-screen reader mode for the rest of the sitting.

In reader mode the learner sees the current question and a small set of large controls — play and pause, repeat, previous and next, speed within the limits the school allows, and progress through the paper. There is nothing else on screen: no papers to edit, no learner records, no admin navigation.

Every playback event is logged locally and synced back to the server, so the session can be reconstructed afterwards if a result is ever queried.

That's it. No complicated setup for learners. No human reader sitting beside them throughout the exam.

Before exam day

We hand schools a deployment checklist. The items that matter most on the day:

• Put the server on a UPS. A power blip mid-exam means a silent reader.
• Disable sleep and hibernate on the server.
• Pause Windows Update for the exam period.
• Add antivirus exclusions for the install and data folders.
• One set of headphones per learner, tested beforehand.