AZMUTH

AZMUTH converts handwritten answer sheets into multidimensional cognitive profiles. Automatically. At scale.

See How It Works →
0 Components
0 Cognitive Metrics
0 Evaluation Rules
0 Purpose
The Problem

A Number Tells You Nothing

StudentScore
Student A 58Identical
Student B 58Identical

Student A — Cognitive Profile

Conceptual Clarity82%
Procedural Accuracy34%
Logical Sequencing71%
Persistence Index88%
Memory Reliability45%
Before AZMUTH
After AZMUTH
📊

Raw Grades

Single-dimensional scores that hide cognitive reality

📝

Generic Feedback

"Needs improvement" — but in what, and why?

⏱️

3–6 Hours Marking

Per class of 40 students — manual, exhausting labor

How It Works

The Five-Stage Pipeline

01
📄

Answer Sheet Upload

Teachers upload photos or scans of handwritten answer sheets. Multi-page support for comprehensive exams.

→
02
🔍

AI Vision Analysis

Advanced AI reads and interprets handwriting — extracting content, structure, and error patterns from every answer.

→
03
🧩

Question Extraction

Each response is decomposed into sub-steps, identifying what the student did — and where the cognitive breakdown occurred.

→
04
🧠

Cognitive Profiling

11 cognitive metrics are computed using the 15-rule classification system. Each student gets a unique cognitive fingerprint.

→
05
📋

Report Generation

Comprehensive cognitive report with archetype classification, causal chains, and actionable pedagogical recommendations.

~30 seconds per student
The Cognitive Engine

11 Dimensions of Student Intelligence

🧠 Understanding
Conceptual ClarityCore
Measures whether the student truly understands underlying concepts vs. surface-level memorization.
Foundational GapCritical
Identifies missing prerequisite knowledge that cascades into downstream failures.
Memory ReliabilityVariable
Detects whether errors come from forgetting formulas vs. misunderstanding them.
⚙️ Execution
Procedural AccuracyCore
Tracks step-by-step execution precision — are the right methods applied correctly?
Logical SequencingCore
Evaluates whether problem-solving steps flow in a coherent logical order.
Verification InstinctVariable
Does the student double-check work? Caught self-errors indicate metacognitive strength.
💡 Thinking Style
Analytical Thinking IndexCore
Measures depth of reasoning — can the student break complex problems into components?
Pattern Recognition StrengthVariable
Tracks ability to identify recurring structures and apply learned templates to new problems.
🔥 Under Pressure
Cognitive Load ThresholdCritical
Identifies the complexity ceiling — where does the student's performance collapse?
Persistence IndexCore
Measures grit — does the student keep attempting difficult problems or abandon them?
Consistency IndexVariable
Tracks performance variance across easy vs. hard questions — how stable is their cognition?
Student Profiles

The 7 Student Archetypes

Causal Intelligence

Why Failures Cascade

Azmuth doesn't just list weak metrics — it maps why failures cascade through connected cognitive dimensions.

Classification System

Subject-specific deterministic rules.
Not guesswork.

azmuth_rules.sys — 15 active rules
[01] IF conceptual_error AND procedural_correct → Foundational Gap detected
Student applies methods correctly but starts from wrong premise — the foundation is rotten.
[02] IF step_skip > 2 AND final_correct → Pattern Matching without Understanding
[03] IF error_consistency > 80% → Systematic Misconception flagged
Repeated identical errors indicate a stable but incorrect mental model.
[04] IF attempt_rate < 40% AND difficulty> median → Cognitive Overload Trigger
[05] IF self_correction_rate > 30% → Strong Verification Instinct
[06] IF variance_across_topics > 60% → Inconsistency Pattern detected
High performance variance across topics signals fragmented understanding, not general weakness.
[07] IF formula_recall_fail > 50% → Memory Reliability Risk
[08] IF logical_order_break > 3 → Sequencing Deficit flagged
[09] IF partial_marks_ratio > 60% → Process-Strong, Answer-Weak profile
[10] IF abandon_rate > 50% AND easy_score > 80% → Persistence Collapse under Load
Student performs well on easy questions but completely abandons difficult ones — a clear cognitive load threshold.
[11] IF similar_q_variance < 10% → High Consistency Index
[12] IF novel_problem_score < 30% → Low Pattern Transfer ability
[13] IF decomposition_rate > 70% → Strong Analytical Thinking
[14] IF careless_error_rate > 40% AND concept_score > 75% → Execution Gap
[15] IF all_metrics > 65% AND variance < 15% → Complete Learner classification
The Roadmap

Azmuth Native AI

Stage 1 — Current

Claude Sonnet

Production-grade AI backbone. Proven accuracy on cognitive profiling.

→
Stage 2 — In Development

Azmuth Native AI

Proprietary model trained on structured cognitive assessment data.

0

structured parameters · per student · per exam

Metric Current (Claude) Native AI Target
Mistake Classification Accuracy ~85% 95%+
Handwriting Recognition ~90% 98%+
Cost Per Analysis ₹8–12 ₹1–2
Latency Per Student ~30 sec <5 sec
Training Data Accumulating
Market & Scale

India's K–12 Opportunity

0 Schools
0 Teachers
0 Students K–12

6-Month Rollout Milestones

Month 1

5 pilot schools
₹1.5L MRR

Month 2

25 schools
₹7.5L MRR

Month 3

100 schools
₹30L MRR

Month 6

1,000 schools
₹90Cr ARR target

Revenue Projection

₹1.5L
M1
₹7.5L
M2
₹30L
M3
₹1.5Cr
M4
₹4.5Cr
M5
₹7.5Cr
M6
In Action

See AZMUTH in Action

Visit YouTube Channel →
Defensibility

Competitive Moat

🛡️

Proprietary Prompt System

22-component structured prompt architecture. Not a ChatGPT wrapper — a precision instrument.

The prompt system encodes pedagogical expertise, subject-specific logic, and deterministic evaluation rules that took months to develop and cannot be reverse-engineered.
📊

Training Data Asset

Every analysis generates 191 structured parameters. This data compounds into an irreplaceable asset.

As more students are analyzed, this proprietary dataset grows — the foundation for Azmuth's own native AI model that no competitor can replicate.
🔄

Teacher Workflow Lock-in

Once teachers build cognitive histories for students, switching costs become prohibitive.

Longitudinal cognitive tracking across semesters creates irreplaceable longitudinal profiles — the more you use it, the more valuable it becomes.
🧬

Identity Graph

Each student develops a unique cognitive fingerprint that evolves over time.

The identity graph maps cognitive evolution across assessments, creating a living profile that no point-in-time assessment tool can match.