Voltage
Marking Maths, Physics & theory today

It marks themethod,not just the answer.

QuestPix reads handwritten answer sheets like an experienced examiner - criterion by criterion, step by step. Correct reasoning earns credit even when the last line is wrong.

Watch it mark a paperBring us one exam
Q7 · Numerical · 5 marksReading
u = 5 m/s, v = 0, t = 4 s✓
v = u + at✓
0 = 5 + a(4)✓
a = −5/4 = −1.20 m/s²✗
F = 2 × 1.20 = 2.4 NECF
C1 · Data extraction-
C2 · Kinematic relation-
C3 · Substitution-
C4 · Retardation value-
C5 · Force from F = ma-
Awarded0.0 / 5Recognising handwriting…
31 hrs
Saved a checker each week
4.2×
Faster than manual marking
61
ECF marks recovered per class
100%
Marks traceable to evidence
Scroll to watch it mark

One answer. Five criteria. One honest score.

The student's page
u = 5 m/s, v = 0, t = 4 s✓
v = u + at✓
0 = 5 + a(4)✓
a = −5/4 = −1.20 m/s²✗
F = 2 × 1.20 = 2.4 NECF
Rubric v1 · Q74.0 / 5
C1 · Data extractionAll four quantities with units1.0 / 1
C2 · Kinematic relationv = u + at - one of three accepted routes1.0 / 1
C3 · SubstitutionConsistent with the student's own values1.0 / 1
C4 · Retardation valueAE - 5 ÷ 4 given as 1.200 / 1
C5 · Force from F = maCredited under ECF1.0 / 1
Criterion 5 - follow-through

F = ma is applied correctly to the student's own wrong value, so ECF credits it. 4 of 5, and because Ce is 0.82 the answer still comes to you to confirm.

The difference

Most tools compare the final answer. That is not marking.

Answer matching
A wrong last line scores zero, however sound the method.
Any method absent from the model answer is rejected.
One arithmetic slip is punished again at every later step.
Theory marking degrades into keyword spotting.
Unclear handwriting silently becomes a wrong answer.
QuestPix
Method, reasoning and accuracy are separate criteria with their own marks.
Alternative valid routes are verified on their own terms.
Follow-through credits correct work on the student's own wrong value.
Theory is read as a concept graph - paraphrase earns full credit.
Mark-affecting uncertainty is escalated to you, never guessed.
How it works

Upload the paper. Approve the rubric. Review the exceptions.

01

Upload the paper

A question paper as a PDF, a photo, a Word file or pasted text. QuestPix splits it into questions, marks and types, and hands you the list to correct.

02

Approve the rubric

Criteria are drafted per question and stay proposals until you lock them. Marks belong to criteria, not to a final answer - and locking creates v1, not a silent overwrite.

03

Review the exceptions

Confident answers are accepted against your own threshold. What reaches you is what could actually move a mark, ordered by how much.

Forty copies a night

The fortieth copy deserves the same examiner as the first.

Consistency is the first thing to go at 11 p.m. QuestPix marks copy forty against the same approved rubric as copy one, then flags equivalent answers that somehow received different marks.

Semantically equivalent answers that received different marks are flagged for moderation.
Approve an alternative method once; equivalent answers follow, outliers stay separately scored.
A high override rate on a question prompts a rubric review instead of hiding in the average.
Consistency audit
Copy 04 · Q74.0 / 5
Copy 22 · Q74.0 / 5
Copy 40 · Q72.0 / 5

Three answers the marker read as equivalent. The third scored two marks lower, so it is held for moderation rather than published.

What is inside
01

Rubric engine

Every subjective question becomes machine-readable criteria: partial levels, dependencies, tolerances, unit rules, significant figures and penalty caps. Edits after checking begins create a new version instead of silently rewriting history.

VersionedAny N of MTeacher authority
02

Step-aware mathematics

Each transformation is validated as valid, invalid, valid-under-condition or uncertain. Symbolic equivalence, domains and assumptions are checked - so 1/√2 and √2/2 are the same answer, and a correct result from invalid reasoning does not quietly earn method marks.

Symbolic equivalenceProof chainsGeometry
03

Physics & chemistry protocols

Data extraction, principle, formula, substitution, sign, unit and precision are assessed separately. Dimensional sanity checking catches student slips and lowers recognition confidence when OCR yields impossible units. Structures compare connectivity and charge, not pixels.

Dimensional checkUnit conversionMechanisms
04

Confidence-based routing

Recognition, interpretation and evaluation confidence are scored separately, and uncertainty is prioritised by whether it can actually move a mark. Thresholds are configuration, never hardcoded into academic logic.

Cr / Ci / CeMarks at riskZero-mark audit
05

Audit that holds up

Source image, recognised content, structured interpretation, rubric version, criterion scores, errors, overrides and publication history are all retained. Every displayed score expands into the reasoning behind it.

Immutable logOverride reasonsRecheck trail
Subject coverage

Each subject is marked by its own protocol.

MathematicsLive

Step validation, symbolic equivalence, proofs and geometry with construction and theorem selection assessed separately.

PhysicsLive

Numericals, derivations and diagrams - circuits, ray and force diagrams with type-specific connection rules.

Theory & long answersLive

Concept graphs instead of keywords. Command words decide what evidence a mark actually requires.

ChemistryPhase 2

Equations, balancing, mechanisms and stereochemistry compared as chemical graphs.

Diagrams & graphsPhase 2

Components, labels and target association, axes, scale and plotting within rubric-defined tolerance.

Mixed languagePhase 2

English, Hindi and Hinglish answers, marked on the same rubric as everything else.

From the marking desk
“It found method marks I would have missed at 11 p.m. on the fortieth copy - and it argued its case with the student's own working.”
R. DeshpandeSenior Physics checker · 14 years · Pune
42copies marked in the pilot, in one evening
9%of AI marks needed a teacher override
61follow-through marks a final-answer check would have discarded
Pricing

Priced per copy checked, not per seat.

Checker
₹8per copy

For an individual teacher running their own exams.

Up to 500 copies a term
Maths, Physics and theory marking
Review queue and evidence highlighting
Checked copies with annotations
Start free trial
DepartmentMost chosen
₹6.5per copy

For a subject department marking together.

Unlimited copies and exams
Rubric library and versioning
Answer clustering and consistency audit
Class, topic and error analytics
Calibration against checked samples
Book the pilot
Institution
Custom

For schools and coaching chains with multiple branches.

Multi-teacher moderation
Admin, roles and batch management
Scanner and LMS integration
Institution-wide analytics
On-premise storage options
Talk to us

Per-copy pricing counts a copy once, however many times it is re-checked after a rubric change.

FAQ

The questions checkers actually ask.

Anything else, we will answer it on a call with your papers open.

Only where you allow it. Answers above your auto-accept threshold are accepted, everything below is highlighted or routed to you, and mandatory review overrides confidence for unusual methods, ambiguous handwriting and rubric conflicts. Publishing is always a deliberate teacher action.

Unclear handwriting creates uncertainty, not a wrong answer. Recognition confidence drops and the item is prioritised by how many marks the ambiguity could move - a doubtful character that cannot change the score stays low priority; an overwritten five-mark answer comes straight to you.

Yes. Alternative methods are verified independently rather than rejected for differing from the model. You can approve a route once and apply it to genuinely equivalent answers, while outliers in the same cluster stay separately scored.

A new version is created. You choose the re-evaluation scope - every answer for that question, only the affected answer patterns, or defer - and earlier evaluations stay associated with the version that produced them, so historical marks can still be explained.

In your institution's own storage tenancy, with source images retained as evidence for every mark. Sheets are never used to train models, and on-premise options are available on the Institution plan.

Bring one exam. We will mark it beside you.

Send a question paper, a marking scheme and forty copies. You will see criterion-level marking, the review queue and a checked copy on your own paper - not a canned demo.

Answer sheets stay yours. Nothing is used to train models.