Repetition takes time.
Teachers repeatedly compare answers, allocate marks, and write feedback across large batches.
SUHI AI · COLLECTIVE INTELLIGENCEAI intelligence.
AI intelligence.
Human judgment.
Every answer matters.
From a handwritten answer to feedback a student can act on.
SuhiMark brings AI-assisted evaluation, subject-expert review, and institutional control into one assessment journey.
A walkthrough for SIH juries · Product vision & evidence in progress
THE COLLECTIVE INTELLIGENCE LOOP
01 / MACHINE INTELLIGENCE
Find the evidence.
Question-wise marks. Rubric-grounded feedback.
AI + HUMAN BRAIN
02 / HUMAN JUDGMENT
Understand the nuance.
Subject expertise. Corrections. Accountable approval.
The institution has the final word.Reviewed first. Published by choice.
01 / THE CHALLENGEA mark closes an exam.
A mark closes an exam.
Feedback opens a possibility.
Checking a paper is more than finding the correct answer. It means understanding a student’s reasoning—and showing them how to improve.
Context changes the judgment.
Partial working, alternative methods, and unclear handwriting need more than a single model response.
Trust needs accountability.
Every proposed mark should be reviewable. Every published result should have a clear approval path.
OUR APPROACH
Let AI do the first pass.
Let educators make it count.
02 / THE ASSESSMENT JOURNEYOne copy.
One copy.
A clear chain of responsibility.
The intended end-to-end workflow: from the institution’s student directory to expert-reviewed feedback and a controlled release.
- 01Institution
One institution. One workspace.
Admins and tutors manage student IDs, batches, the master paper, model answers, and the marking rubric.
OUTPUTStudent directory + assessment rubric
- 02Institution
Every copy has an owner.
Scan and upload each answer sheet against the correct student ID. Unclear handwriting needs confirmation before it can support a mark.
OUTPUTStudent-linked scan + confirmed transcription
- 03SuhiMark · Phase 1
AI prepares the first evaluation.
Questions are evaluated individually against the rubric. The proposed checked report pairs answers with marks, strengths, and specific improvement feedback.
OUTPUTQuestion-wise draft evaluation
- 04SuhiMark · Phase 2
Experts question the evaluation.
Subject experts inspect the evidence, discuss the assessment, adjust marks or feedback, and approve the reviewed copy. A discussion alone does not change a grade.
OUTPUTExpert-reviewed marks + revision history
- 05Institution
Institutions see the complete picture.
After expert approval, the institution receives the checked copy. Teachers can request refinements from the SuhiMark review team before publication.
OUTPUTChecked report + clarification requests
- 06Institution → student
Release remains an institutional decision.
The institution publishes the result. Students can then access feedback; parent delivery follows the institution’s policy and explicit approval.
OUTPUTPublished result + approved parent delivery
A checked report, question by question. Each report is designed to bring the question, student answer, awarded marks, strengths, and improvement feedback together. Expert corrections remain part of the review history.
03 / UNDER THE HOODBuilt around trust.
Built around trust.
Designed to grow.
Explore the reference architecture. Each stage explains its inputs, outputs, controls, and current implementation status.
REFERENCE ARCHITECTURESelect a stage to explore
Shared data layerAssessment records·Job queue·Review revisions·Audit eventsPostgreSQL foundations
Inside stage 03
AI evaluation
A model adapter requests structured rubric evidence and feedback. Validated criterion awards determine the total. Uncertain outputs are referred for human review.
- Input
- Answer + question + rubric
- Output
- Proposed marks + grounded feedback
ControlModel output is untrusted. Validation precedes any saved evaluation.
Gemini adapter and validation foundations; OCR integration and benchmark pending.
One clear interface.
Next.js and TypeScript connect institution workflows with review experiences.
Work continues asynchronously.
A durable SQL queue, leased jobs, and guarded retries isolate model processing from user interactions.
A revision, not a silent overwrite.
Revision checks, audit events, and separate approval gates protect the evaluation lifecycle.
What is implemented, and what is still being validated?
Available foundations: public website, sample review experience, database schemas, evaluation worker, model adapter, review transactions, PDF generation, and delivery queue logic.
Before live assessment operation: production identity, scan/OCR and object-storage integrations, complete persistent UI wiring, delivery adapters, measured accuracy, security verification, and load testing.
Capacity: 10,000+ students is a V1 design target. It is not a verified capacity or a claim of current usage.
04 / MEASURE WHAT MATTERSEvidence before promises.
Evidence before promises.
Accuracy, made accountable.
Benchmark results will be added with sample sizes, evaluation conditions, and expert reference scores. A single percentage cannot tell the whole story.
Handwriting recognition
—Character / word error rateCompare confirmed transcriptions with independently labelled answer sheets. Report results by language, scan quality, and subject.
Marking agreement
—Agreement with expert consensusMeasure question-level agreement and absolute mark differences using a held-out set and an agreed scoring tolerance.
Human review impact
—Corrections & review timeRecord how often experts change marks or feedback, plus time spent reviewing each paper.
Operational reliability
—Completion, latency & costReport completed jobs, retry rates, end-to-end turnaround, and cost per reviewed copy under measured load.
PILOT & FIELD LEARNINGFrom real classrooms
From real classrooms
to meaningful evidence.
Working with 5+ institutions, as reported by the founding team. This reflects institutional relationships, not a published outcome study.
Details to follow
Pilot scope
Institution names, dates, subjects, and consented cohort details
Details to follow
Measured outcomes
Copies reviewed, grading agreement, review time, and cost
Details to follow
Lessons learned
Observed limitations, educator feedback, and next changes
SEE THE PRODUCT DIRECTIONMake the process
Make the process
tangible.
A product demonstration of the assessment experience. This video illustrates the product direction; implementation status is documented above.
05 / THE PEOPLE BEHIND THE SYSTEMDifferent disciplines.
Different disciplines.
One shared responsibility.
Meet the team here soon. Names, photographs, credentials, and individual contributions will be added after confirmation.
Product & institution partnerships
Classroom needs, pilot design, and institutional adoption.
AI & assessment engineering
Document understanding, rubric evaluation, and measurable reliability.
Subject expertise & review
Academic judgment, marking consistency, and useful feedback.
Platform & experience
Secure workflows, dependable operations, and accessible interfaces.
WHAT COMES NEXT
Close the loop between assessment and everyday learning.
On the horizon · Coming soonToday’s lesson.
Today’s lesson.
Tomorrow’s confidence.
Turn each day’s learning topic into a focused practice quiz. A small check-in to help students discover what clicked—and what needs another look.
- Choose the topic
- Teacher reviews the quiz
- Students practise
THE DAILY CHECK-INConcept preview
What did you learn today?
BiologySample question / 01
Where does the oxygen released during photosynthesis come from?
- ACarbon dioxide
- BWater
- CGlucose
THE IDEA IS COLLECTIVE. THE IMPACT IS PERSONAL.Better evaluation.
Better evaluation.
More room to learn.
Explore the product, ask about the architecture, or discuss an institutional pilot.