Synque·LMS

Hong Kong secondary · English writing, reading, listening & speaking

AI marking your teachers can check.
And we can prove it.

Every AI mark arrives with the words that justify it. Nothing reaches a student until a teacher releases it. And every capability is measured before it ships — with the results published.

See the live evidence Try the teacher workspace
Proposal: a school repair clubChan Tai Man · 4A

Every week, many students throwsthrow away umbrellas and bags that could easily be fixfixed. A repair club would reduce this waste and teach useful skills. Senior students would teach younger members, and a teacher would supervise each sessionssession.

Every quote checked against the essay

Content5/7
Language6/7teacher: 5 → 6
Organisation5/7
Released by teacher
100%reading marks matched the key
7/7planted errors found, every run
0false corrections on clean writing
241automated checks passing

Real output · eval run, 4 Oct 2026

  • correctionless things fewer things · Use 'fewer' rather than 'less' before countable plural nouns like 'things'.
  • readingThe passage states that 200 passengers were surveyed, not 100.
  • correctionby themself by themselves · Use the plural reflexive pronoun 'themselves' to refer back to 'students'.
  • practiceThe broken printer can easily be ___ by the technician this afternoon.
  • correctionwould reduces would reduce · Modal verbs like 'would' must be followed by the base form of the verb.
  • readingHalf of the passengers travelled every weekday, not on weekends.
  • correctionmore green greener · Single-syllable adjectives usually take '-er' to form the comparative.
  • practiceTaking the MTR during rush hour would ___ our travel time significantly.
  • correctionmake friend make friends · The standard phrase uses the plural form: 'make friends'.

Try the marker

Pick a student’s answer. See what the AI did with it.

Real answers and real marks from our evaluation run — including a misspelling and an attempt to cheat.

What percentage of participants felt confident repairing a button?

“Three fifths of them, which is 60%.”

1AI mark
Its reason
Correct percentage identified.
The passage
A survey found that 18 of the 30 participants felt confident repairing a loose button afterwards.
Marking point
60% (18 of 30)

For teachers

Marking you can check in seconds, not redo in hours.

Writing, marked with evidence

Content, Language and Organisation on an HKDSE-style rubric — each score backed by quotes from the essay, corrections shown in the student's own text. One revision included.

Reading, marked line by line

Paste a passage, write the questions and your marking points. Every answer is marked right or wrong, with the passage line that decides it.

Practice from their own mistakes

Exercises built only from each student's verified corrections. The server marks them, not the AI.

Planning help that never writes the essay

Questions, ideas and phrases before the first draft. Capped in length; anything essay-like is withheld.

The teacher decides

Change any mark, add a comment, release. Every change is kept.

Four skills

One standard for writing, reading, listening and speaking.

Upload a recording; the AI drafts the transcript, the teacher approves it. Students answer by ear; the transcript stays hidden until release.

Each mark quotes what was said:

  • Where is the workshop now?“library”
  • What time does it start?“three thirty”
  • How many places are there?“twelve places”

Promise 1, in your hand

What the student sees.

The AI has marked the essay. Switch between before and after the teacher releases it.

Chan Tai Man · 4A9:41

Proposal: a school repair club

Submitted. Your feedback appears here when your teacher releases it.

Enforced, not promised

Four promises the system keeps

Each is guarded by an automated test that we broke on purpose to prove it catches the failure.

  1. 01

    Nothing before release

    No mark, feedback or answer key reaches a student until the teacher releases it — checked against what the server actually sends.

    $ npx tsx test-reading-service.ts

    ✓ RD1 the task list a student receives has no marking key

    all passed

  2. 02

    No claim without evidence

    Every quote the AI relies on is matched against the essay or passage. One that is not really there is hidden and flagged.

    $ npx tsx test-reading-service.ts

    ✓ RD3 a passage quote that is not in the passage is withheld

    all passed

  3. 03

    The numbers add up

    Totals are computed by our code, never copied from the AI. An essay credit is used once — never for a failed or repeated submission.

    $ npx tsx test-reading-service.ts

    ✓ RD2 the total is the server's sum (2), not the model's claimed 3

    all passed

  4. 04

    Only your own work

    Students see only their own work. Teachers see only their own classes.

    $ npx tsx test-reading-service.ts

    ✓ RD5 a student in another class sees no task

    all passed

Measured, not promised

The scorecard

The product's own marking code, run against cases whose answers were written before the AI saw them.

Our cases, not yet your teachers’. The pilot measures agreement on essays your teachers have already marked.

Reading answers matched the marking key38 answers × 3 runs — paraphrased, misspelled, near-miss, blank100%
Marks wrongly givenincluding two attempts to tell the marker to give full marks0
Listening answers marked from a real recording45 marks across three cases; transcript word-for-word100%
Planted grammar errors foundevery run, each with the right correction7 / 7
Corrections on an error-free letterthe AI leaves correct writing alone0
Same essay, marked againtotal out of 21, three runs per essay±1
Automated checkspublished on the evidence page241 / 241

Getting more accurate

Every mistake becomes a test it never fails again.

  1. 1

    Measure

    Run every case. Score it.

  2. 2

    Find the misses

    Read each wrong answer and why.

  3. 3

    Change one thing

    Instructions, checks, context or model.

  4. 4

    Measure again

    Keep it only if accuracy rises and nothing falls.

Every change a teacher makes to an AI mark is already recorded. In the pilot those changes become test cases — so the AI is measured against your own teachers’ standard.

Synque

How we build AI

Rules before code
What must never go wrong is written first, as a numbered rule that the code and its checks both cite.
Checks that can fail
A new check is broken on purpose before we trust it. A test that has only ever passed proves nothing.
People release decisions
AI proposes; a person approves. Here, no AI mark reaches a student until a teacher releases it.
Evidence anyone can open
Journeys, screenshots and check results on one page — including what is not yet measured.

Bring 30 marked essays. We’ll show you the number.

Before any student uses it, see how closely the AI agrees with your English panel.

Email us WhatsAppSee the live evidence

HKDSE-style rubric for formative feedback — not official HKEAA bands. Figures measured 4 Oct 2026 on gemini-3.8-flash.