Administrator
Sees attainment across programmes and departments, programme accreditation status, grading parity between sections, and institution-level integrity indicators.
The platform
Everything it takes to run an exam whose result you can defend
An item bank, a delivery engine, proctoring, scoring and analytics — in one system. The difference is that each layer feeds the next, so an exam ends in evidence rather than a grade.
Arabic-firstHosted in-KingdomAccreditation-alignedContinuous psychometricsAI with stated limits
Six capabilities, each solving a problem the department knows
Select any one to see exactly what it covers. The order is not arbitrary: the output of each is the input of the next.
An unreviewed item in the second corrupts the analysis in the sixth — which is why none of them is sold on its own.
01
Generate items from your own content, and measure what multiple choice cannot
From a course file to an outcome-mapped draft bank — then item types that reach past recall.
- Item generation from an uploaded file or pasted text, at a set difficulty and outcome
- Proposed scoring rubrics for essay items, edited by the instructor
- Mathematical, chemical and engineering notation with an on-screen keypad
- Coding items that actually execute in an isolated sandbox
- Generated values that differ per student, so a copied answer is wrong
- An editor that handles Arabic prose and Latin code in the same block
01 / 06
02
An item bank that neither bloats nor leaks
A relational structure tying every item to an outcome — not a free-text field.
- A personal bank per instructor and a shared institutional bank with defined permissions
- A mandatory lifecycle: draft → pending review → approved → archived
- Preserved versions, so a later edit never corrupts an earlier report
- Bloom's level and learning domain inherited from the outcome
- Duplicate detection on import with correct Arabic normalization
02 / 06
03
Exam day passes without anything breaking
A different form for every candidate, at an equivalent difficulty weight.
- Randomized draws from a larger pool with shuffled options
- Timers at the assessment, section and item level
- Adaptive delivery on an IRT engine with exposure control
- Offline-first operation that reconciles when the network returns
03 / 06
04
Integrity through reviewable evidence, not automated verdicts
Human-reviewable evidence — not automated verdicts.
- Documented alternative verification paths where facial matching does not apply
- Each monitoring layer enabled independently, per assessment
- A timestamped snapshot and event log attached to every flag
- A measured false positive rate, computed from reviewer dispositions
04 / 06
05
Instant where it can be, human where it must be
Automated where it can be, human where it must be.
- Instant scoring for closed item types
- Calibrated tolerance for minor spelling variation in short answers
- A mathematical equivalence engine that knows x+y equals y+x
- LLM-assisted essay evaluation against a rubric, with the final call human
05 / 06
06
From a grade to an action whose effect is measured
From a grade to an action whose effect is measured next cycle.
- Attainment reports at both course and programme outcome level
- A course quality report articulating strengths and areas for improvement
- Role-tailored dashboards: administrator, instructor, student
- Actions recorded with an owner and a date, then re-measured automatically
- Export ready for the accreditation file, documented with threshold and method
06 / 06
Three roles, three different screens
Not one dashboard for everyone. Each role sees what it needs to decide, and nothing that is not its business.
Instructor
Generates and reviews items, builds rubrics, approves proposed scoring, and sees the quality of their own items before anyone else does.
Student
Sits the assessment, sees their position against outcomes rather than a bare grade, the topics needing review, and their verifiable credential.
AI at eight points, each with a boundary
Not a bolted-on layer but capabilities working inside the six above — starting with generating items from your own content.
Generating items from your own content
Upload a course file or paste text, set the difficulty level and the target outcome — and the system generates items already linked to that outcome, landing in the draft bank for review rather than use.
Building scoring rubrics
For essay items the system proposes rubric criteria and performance levels from the item text and its linked outcome, which the instructor edits before approving.
Adaptive delivery
An item response engine re-estimates ability after every answer and draws the most informative item at that level, reaching the same precision with fewer items.
Identity verification and proctoring
Continuous facial matching, detection of multiple faces, sustained gaze deviation and human speech, plus clustering that surfaces identical wrong-answer patterns across candidates.
Rubric-based essay scoring
The system evaluates an essay against the approved rubric and proposes a score with reasoning per criterion — and no score is finalised without instructor review.
Quality reports
The system generates a course quality report articulating strengths and areas for improvement from attainment and item analysis data — in the form the course report asks for, not as a table of numbers.
Early warning
A classification model reads historical performance, response time and completion rate to estimate the probability of a student falling behind — early enough to intervene.
Generated executive summaries
Turning standard deviations and attainment percentages into a readable paragraph explaining what happened, and for whom it matters.
Three things AI does not decide in EvaliX
A decision that cannot be attributed to a person cannot be defended to a student who objects, or to a reviewer.
It does not rule on integrity
Detectors gather timestamped evidence and order the review queue. Any action affecting a student requires a recorded human review, and we measure the false positive rate per detector.
It does not finalise a grade
On essays the model proposes and explains a score; approval stays with the instructor. The approved grade carries the name of whoever approved it in the audit trail.
It does not retune its own thresholds
Changing event severity weights or detection thresholds is a documented decision with human approval. A system that retunes its own integrity thresholds is indefensible in an academic appeal.
The foundation: infrastructure and compliance
All six rest on one foundation. These are not features added later but architectural decisions taken before the first line of code.
Data residency
Databases and proctoring media inside the Kingdom — a default, not a hosting option.
Arabic-first
Full layout mirroring, with Arabic fonts embedded in certificate and report generation.
Immutable records
An audit trail for every change to a grade or approved item; soft deletion only.
Finals-day resilience
Thousands of concurrent sessions, per-interaction persistence, offline-first architecture.
Three differences worth scrutinising
Do not take these on trust — ask us to demonstrate each one.
The closed quality loop
Every analytics dashboard ends at a number. EvaliX records the action taken in response — with a named owner and a date — then re-measures the same metric next cycle. An external reviewer never asks what your attainment was; they ask what you did about it and whether it worked.
Arabic as a first design language
Cosmetic support translates the interface and leaves data tables left-to-right and certificates with disconnected glyphs. In EvaliX Arabic is the first language: full mirroring, an editor that handles Arabic prose alongside code, and fonts embedded in the document engine.
Data inside the Kingdom
Assessment data is personal and proctoring data is biometric. Hosting it abroad turns a technical decision into regulatory exposure — which is why in-Kingdom hosting is the default in EvaliX, not a plan you buy.
An assessment platform or the quiz module in an LMS?
The question is not which is better but what each was built to do. In most large institutions the right answer is integrating the two, not replacing one with the other.
| Dimension | LMS quiz module | EvaliX |
|---|---|---|
| Designed for | Content and activity management; quizzing is secondary | Measurement itself |
| Item types | Mostly closed formats | Code execution, equivalence, generated values, projects |
| Item quality | A manual difficulty tag | Facility and discrimination computed from live data |
| Outcome mapping | A text field or a tag | A database relation with versioning and weights |
| Integrity | A password and a time window | Independent layers, alternative verification, timestamped evidence |
| Accreditation reporting | Assembled by hand at year end | A natural byproduct of daily operation |
See all six on a course from your own institution
A tailored demonstration, not a canned deck. We take a real course and walk it from the item bank to the attainment report.