Skip to content
AI Math Solver 2.0

Marks the first wrong line in each student's working, and says why.

For maths teachers in Malaysian lower secondary schools.

Student A

Solve for x.

  1. 3x + 5 = 20
  2. 3x = 20 + 5First wrong lineMoved a term to the other side without changing its sign
  3. 3x = 25Follows from the wrong line
  4. x = 25/3

Practice set: moving terms across the equals sign

A marked worksheet. Made by computer, not a student's script.

Try the demo dashboardWatch the 2 min video

One class of 30 computer-made worksheets, all checked in 79.1 s (1.3 min) on the live dashboard. See every result

The teacher scans a class set of worksheets. Each answer comes back with the wrong line in red, the misconception named, and practice for that student.

Software prototype running now. The classroom box and school trials come next.

Malaysian Grade 8 maths is slipping

In TIMSS 2023, Malaysia's Grade 8 mathematics score fell to 411, down from 461 in 2019, and close to the Low International Benchmark of 400.

380400420440460480Low benchmark20192023

Teachers rarely have time to read every student's working to find the pattern behind wrong answers. So remedial lessons repeat the whole topic.

Source: TIMSS 2023 international results, Grade 8 mathematics.

Who it is for

Built for the people who mark a class set and plan the next lesson.

Maths teachers.
Lower secondary, marking a full class set at a time. Topics follow the KSSM syllabus, and the teacher picks them.
Heads of maths.
See which misconceptions repeat across classes, and plan remedial lessons together.
Schools.
Worksheets stay on the box, inside the school.

What it does with one stack of worksheets

Each step below runs today in the software prototype, on a demo class of made-up students.

  1. The teacher scans the whole class set in one go.

    Teacher dashboard, demo class: the Integer question with 24 of 24 sheets marked.

  2. In each answer, the first wrong line is marked in red.

    One sheet from the demo class: the second line is circled in red as the first wrong line, with the misconception named beside it.

  3. The misconception behind it gets a name, in Malay or English.

  4. A class map shows which misconceptions are most common.

    Class map of the demo class: one row for each misconception, one column for each student.

  5. Students who share a misconception are put in one group.

    Groups in the demo class: students who share a misconception are listed together.

  6. Each group gets a printed practice sheet on the step they got wrong.

    Printable practice sheet for the group that subtracts a negative number wrongly.

Measured so far

Each result comes with its count and its 95% range. Sets made by computer and real handwriting are kept apart.

Made by computer

One class, timed on the live dashboard

30 computer-made worksheets uploaded at once, as a teacher would. Timed from upload until the last result shows on screen. This ran on the server. The box is not timed yet.

MeasureResult
Whole class done79.1 s (1.3 min)
Worksheets checked30 / 30
Passed to the teacher0 / 30

Made by computer

The whole chain, on scripts made by computer

300 worked answers drawn by computer in a handwriting style, not written by pupils. They show the chain runs from photo to practice set. They do not measure accuracy on pupils' handwriting.

MeasureResultCount95% range
First wrong line found100.0%300 / 30098.7% – 100.0%
Misconception named correctly100.0%204 / 20498.2% – 100.0%
Correct scripts marked wrong (lower is better)0.0%0 / 960.0% – 3.8%

Real handwriting

Reading real handwriting by adults

200 handwritten maths expressions by adults, from a public set. Not pupils, not worksheets. All three results are shown together. We are cutting the misreads without a warning before any school trial.

MeasureResultCount95% range
Read exactly right47.0%94 / 20040.2% – 53.9%
Passed to the teacher to check50.5%101 / 20043.6% – 57.4%
Misread without a warning (lower is better)23.0%46 / 20017.7% – 29.3%

Not measured yet: real students' worksheets. None are collected before research approval, the state education department and parents agree.

Where it stands today

  • Running now: the software prototype and the teacher dashboard, with a demo class of made-up students.
  • Not yet: the classroom box itself, and any trial in a school. Trials start only after research approval.

Student data stays in school

Worksheet images stay on the box inside the school. Student names are blurred on every image. No real student's work appears on this page.

Video and poster

Watch the video (2 min, English, captions)

Who builds it

Inventor: Aiman Malib

Sasa Teguh, the team behind AI Math Solver, gold medal at IPITEx 2020. In 2020 it solved questions. In 2027 it shows why a student went wrong.

AI Math Solver won gold at IPITEx 2020.