Skip to content
MNT
LAYER 1Subject knowledge engineering · Solves inefficient paper review

Turn your strongest subject’s papers and teaching experience into a proprietary knowledge asset.

A barrier no competitor can take from you. The asset belongs to your institution and is never used to serve another client.

Positioning

Not “question data entry” — but turning exam papers into knowledge assets.

Subjects deliveredChinese LanguageEnglish LanguageMathematicsEconomicsBAFS

The engineering capability is subject-agnostic — all subjects and year levels are open to discussion.

Layers of processing

Four layers, breaking a paper down to a granularity a machine can diagnose.

Every layer has a tutor review gate. Pricing follows depth, from baseline processing to full knowledge engineering on a single subject.

LAYER 01

Paper structuring

Structured by year, paper, question type and marking basis, so every question is searchable and assemblable.

LAYER 02

Language granularity

Broken down to chunks, situational expressions and distractor logic — this layer is where error-cause precision comes from.

LAYER 03

Ability mapping

Each question tagged with the ability level it actually tests and its likely error causes, mapped to the official framework.

LAYER 04

Variant generation

Progressive practice items per ability point, so remediation has material — no more blindly redoing whole papers.

/ 01

Worked example: DSE English

The usual approach

Load past papers into a bank, sort by year and question type, let tutors pull questions.

When a student gets it wrong, the system only knows the question was wrong. Working out why still falls to the tutor, paper by paper.

Knowledge engineering

Break down the chunks and fixed collocations behind each question, the synonym substitutions, the inference logic, and how each distractor was mis-designed.

Then map it to HKEAA ability levels — when a student misses a question, the system can tell whether it is vocabulary, comprehension or answering strategy, and pushes variant practice on that same ability point.

/ 02

What we deliver, what we don’t

See the boundary before you buy. The left column works as-is as a contract schedule of deliverables.

We deliver
Structured question bank and subject tagging systemSearchable and assemblable on any dimension, maintained by your own tutors without an engineer present.
Ability-point and error-type mapWhen a student answers wrong, the system identifies which class of error it is — not just right or wrong.
Variant practice setsProgressive items matched to each ability point, for remediation.
Ability radar chart and student report templatesOutput formats a tutor and a parent can both read, badged under your institution.
Teacher handbook and bank-update protocolHow new papers enter the bank and get reviewed — you can run it yourself after handover.
Ownership of the assetThe processed asset belongs to your institution and is never used to serve another client.
We do not deliver
An off-the-shelf question bankWe don’t sell ready-made questions. The raw material is your papers; so is the asset.
AI-invented exam questionsNo model conjuring questions from nothing. Variants derive from ability points and the official framework, then pass tutor review.
Answers generated on the flyAnswers come from official marking schemes, not from a live conversation.
A replacement for tutor markingThe system handles what can be formalised; ambiguity goes back to the tutor.
/ 03

What the diagnosis looks like

The table below is a structural illustration of the output fields. Real mastery metrics and sample data are filled in after your pilot. Error classification is the point — it decides what gets practised next.

Knowledge point
Status
Error class
Suggested next step
Classical function words
Needs work
Knowledge gap
Drill the distinction first, then retry same-type items
Main-idea summary
On target
Maintain
One item every two weeks to hold it
Rhetorical device ID
Partial
Misreading the prompt
Practise circling stem keywords first
Open-response structure
Needs work
Missing strategy
Apply the three-part template; tutor models it once
Register judgement
Undetermined
Undetermined
Returned to the tutor for human judgement

The last row is deliberate: when confidence is short, the system outputs “undetermined” and hands the item back — it never guesses a score and charges it to the student.

/ 04

Marking rules

All three are already in the code. Institutions can read them as contract terms.

The same paper, marked a hundred times, gives the same result.Marking runs on a deterministic rule engine rather than fresh inference each time, so results are reproducible and auditable.
Model answers come from official marking schemes, not generated on the spot.Every answer carries a source marker; tutors and parents can trace it to the official passage.
When it can’t be judged, the output is “undetermined” — never charged to the student.Below the confidence threshold the system stops and returns the item. Better to judge less than to judge wrongly.

Anchored to: EDB curriculum guides · HKEAA marking schemes, all traceable to the official source. Marking precision is promised only where official anchoring exists.

Once the asset exists

Bring one past paper. We’ll take it apart on the spot.

Forty-five minutes. We break down the paper you bring and you watch it become a diagnosable knowledge asset — then we map an entry tier for your institution.

Book a free 45-minute consultation