// EVALUATION.INFRASTRUCTURE
AI Exam Evaluation System
The evaluation layer that turns every exam into structured learning intelligence — built for platforms and institutions processing thousands of answer sheets at scale.
// FIELD.VALIDATION
// THE.SYSTEM
Evaluation infrastructure, not a grading tool
Most platforms evaluate and stop. Scores are recorded. Learning signals — the gaps, misconceptions, step-by-step reasoning — are lost.
Full Pipeline Control
Configure rubrics. Submit answer sheets. AI evaluates. Governed publish. Every step in one system.
Every Modality, Same Signal
Handwritten PDFs, typed text, diagram images — one pipeline. Same structured output regardless of format.
Infrastructure, Not a Tool
Every score arrives with the signals that produced it — feeding personalization, content generation, and learning memory downstream.
// CAPABILITIES
What evaluation infrastructure actually does
Six capabilities, one control point. Every evaluated answer sheet produces the same structured output — regardless of platform, modality, or scale.
Multi-Modality Signal Capture
Handwritten PDFs, typed text, diagram images — one pipeline. Same structured signal output regardless of submission format.
Rubric-Governed AI Evaluation
Define criteria, assign weights, specify step-wise marks. AI evaluates against your rubric — not a generic scoring model.
Two-Stage Publish Control
Pre-publish → final-publish enforced at the API level. No result reaches students without a human approver. Bypass is structurally impossible.
Student Query Resolution
AI responds first — 90%+ accurate feedback delivered instantly. Teacher reviews and approves before the resolution reaches the student.
Embeddable Evaluation Interface
One API call generates an iframe link. Upload sheets, access model answers, view results — no custom evaluation UI build required.
Structured Learning Intelligence Output
Total score, section summaries, step-wise marks, concept-level gaps. The signal layer that feeds personalization and content generation downstream.
// HOW.IT.WORKS
From exam submission to structured learning intelligence
Configure
Create the exam. Attach the model answer with a configurable rubric and step-wise marks per question. The configuration defines what the AI evaluates against.
Capture
Submit student answer sheets via API or embeddable UI. Handwritten PDFs, typed text, and diagram images all accepted. Async webhooks confirm processing at each stage.
Evaluate
AI evaluates every sheet against the configured rubric. Step-wise marks assigned. Question-level feedback generated. Provisional scores assembled with concept-level signals — ready for human review.
Govern and Output
Provisional results enter governed review. Students raise queries — AI responds first, teacher approves. Final scores publish with full structured output and persist as learning signals.
# Step 1: Create exam and configure model answer
POST /evaluation-platform/v1/exam
→ { "examId": "ex_847291" }
POST /model-answer-sheet
PATCH /model-answer-sheet/{id} ← upload PDF/JPEG with rubric
# Step 2: Submit student answer sheets
POST /student-answer-sheet
PATCH /student-answer-sheet/{id} ← upload scanned doc
# Step 3: Retrieve provisional scores
GET /provisional-scores
# Step 4: Governed publish and final output
POST /publish-result ← pre-publish → final-publish
GET /final-scores/{answer_sheet_id}// WHY.THIS.WORKS
Validated at Production Scale. Built to Be Stricter Than a Teacher.
// RUBRIC.ADHERENCE
3.91
marks stricter than teachers, on average
The gap is not error. It is consistency.
// QWK.STANDARD
1.000
median Quadratic Weighted Kappa
The psychometric gold standard. Met at the median question.
// INFRASTRUCTURE.GRADE
99.6%
question-to-rubric mapping success
Evaluation accuracy starts with pipeline reliability.
HUMAN.IN.THE.LOOP
Policy-aware. Human-verified. Infrastructure-grade.
CGF's evaluation system is human-in-the-loop by design. Teachers remain the control point — AI handles volume, humans handle governance. This is not an automation layer. It is evaluation infrastructure with built-in human authority at every control point.
// GOVERNANCE
AI-first, human-controlled at every stage
Pre-publish review is a hard gate built into the system architecture. Governance is structural, not optional.
Two-Stage Publish Gate
Pre-publish → final-publish is a mandatory workflow enforced at the API level. No result locks without a human approver. Bypass is structurally impossible.
Teacher-Reviewed Query Resolution
Student query → AI response → teacher approval. Three parties. The AI is the first responder, not the last word.
Full Audit Trail
Every evaluation action, score override, and query approval is logged. The isApproved field in every final score payload is the API-level record that human review occurred.
// APEX.CONTROL
HUMAN.IN.THE.LOOP
Every critical decision point has a human in the chain — not as an override option, but as a structural requirement enforced at the API level.
// BUILT.FOR
Four operations that need evaluation infrastructure
EdTech Platforms
You evaluate at scale but only capture scores. The Exam Evaluation system adds a structured signal layer to your platform — without rebuilding your evaluation stack.
Coaching & Test-Prep Networks
Subjective evaluation drifts across large cohorts. Rubric-governed AI evaluation eliminates evaluator inconsistency. 93% reliability, auditable at every step, consistent across every cohort.
LMS / ERP Platforms
Your platform generates scores, not structured learning intelligence. Embed the evaluation layer via API or embeddable UI. No custom evaluation UI build required.
Institutions & Public Programs
Manual evaluation at institutional scale is slow and produces no learning memory. Governed AI evaluation with full audit trail — human control points preserved, policy-aware from configuration.
// API.LAYER
A complete evaluation API built for production
REST API
OAuth 2.0 JWT Bearer authentication. Every evaluation step is an API call.
Full Pipeline Control
Create exam, configure model answer, submit student sheets, retrieve provisional scores, manage queries, publish results.
Async Webhooks
Dedicated webhook templates for question extraction, model answer extraction, and answer sheet evaluation events.
Embeddable UI
One API call generates an iframe link — upload sheets, access model answers, view results. No evaluation UI build required.
{
"answer_sheet_id": "as_847291",
"exam_id": "ex_123456",
"student_id": "std_00391",
"isApproved": true,
"total_score": 78,
"max_score": 100,
"section_summaries": [
{
"section": "Organic Chemistry",
"score": 32,
"max": 40,
"signals_captured": true
}
],
"step_wise_marking": [
{
"question_id": "q_01",
"awarded": 6,
"max": 8,
"feedback": "Mechanism shown — final step incomplete"
}
],
"system_stores": "LEARNING_SIGNALS"
}// USE.CASES
See how it works in your context
EdTech Platforms
You evaluate at scale. See how evaluation infrastructure captures the learning signals your platform is currently losing.
See EdTech solution →
Coaching & Test-Prep Networks
Subjective grading needs consistency. See how rubric-governed evaluation eliminates drift at cohort scale.
See Coaching solution →
LMS / ERP Platforms
Your platform manages learning. Add the evaluation layer that produces structured signal output.
See LMS solution →
Institutions & Public Programs
Scale without losing control. Governed evaluation with full audit trail for institutional and public program needs.
See Institutions solution →
// COMMON.QUESTIONS
AI exam evaluation converts exam answer sheets into structured learning intelligence — not just scores. CrazyGoldFish processes every submission, captures concept-level signals, and stores them as persistent learning memory. With 93% scoring reliability (ICC) and a governed human-in-the-loop review workflow, it is evaluation infrastructure, not a grading tool.
What is CrazyGoldFish's Exam Evaluation system?
Evaluation infrastructure for education. It handles the full pipeline — exam creation, model answer configuration, student submission, AI evaluation, governed review, and final score publication — and produces structured learning intelligence at every step. Not a scoring tool. The evaluation control point.
How does the Exam Evaluation system handle subjective answers?
Rubrics are configurable: define criteria, assign weights, specify step-wise marks per question. The AI evaluates each response against your rubric — not a generic model. Every answer produces step-wise marks, question-level feedback, and concept-level signal output. Handwritten, typed, and diagram submissions all use the same pipeline.
Is the Exam Evaluation system a tool or infrastructure?
Infrastructure. A scoring tool returns a mark. CrazyGoldFish's Exam Evaluation system returns structured learning intelligence — step-wise marks, concept gaps, feedback — that feeds personalization, content generation, and learning memory downstream. It is the control point, not a standalone application.
What accuracy does the Exam Evaluation system achieve?
93% reliability in total scoring alignment (ICC). In independent adjudication, AI evaluation achieved 89.7% adjudicated correctness versus teacher evaluation at 82.8% — a +6.9 percentage point improvement. 87% of AI scores fall within ±6 marks of the human benchmark. Published: India AI Impact Summit 2026 Compendium, MeitY.
How does the Exam Evaluation system integrate with an existing platform?
Two paths. REST API with OAuth 2.0 JWT authentication gives full programmatic control over every pipeline step, with webhooks for async evaluation events. Embeddable UI requires a single API call to generate an iframe link — no evaluation UI build required on your side.
// RELATED · RESEARCH & PLAYBOOKS
// NEXT.STEP
The evaluation layer your platform is missing
Every exam your platform runs is producing learning signals you are not capturing. CrazyGoldFish's Exam Evaluation system is the infrastructure layer that changes that.