TECHNICAL ARCHITECTURE SUMMARY

 

PART 2: TECHNICAL ARCHITECTURE SUMMARY

AI-Generated Dynamic MCQ System for NEET

Prepared For	NTA Technical Teams
Classification	CONFIDENTIAL
Document Length	Summary version (full version available separately)

 

SECTION 1: SYSTEM ARCHITECTURE OVERVIEW

HIGH-LEVEL FLOW

NTA COMMAND CENTER (Secure)

  ↓

  - Exam Management Console

  - AI Engine Server

  - Key Management Service

  ↓

EXAM CENTERS (Distributed - 5,400 total)

  ↓

  - Biometric Gate (Face + Fingerprint + Voice)

  - Terminal PC + Offline Question Vault

  - Local Decryption & Delivery

  - Student Response Collection

  ↓

Sync Back to NTA (Encrypted)

SECTION 2: AI QUESTION GENERATION ENGINE

INPUT SPECIFICATION

{

  "exam_id": "NEET_2027_Q1_SESSION_A",

  "subject": "Chemistry",

  "chapter": ["Organic Chemistry", "General Organic", "Hydrocarbons"],

  "difficulty_level": 6,

  "question_count": 60,

  "language": "English",

  "timestamp": "2026-06-21T08:45:00Z",

  "center_id": "NCR_001",

  "candidate_id": "HASH_ANON"

}

PROCESSING STEPS

Step 1: Template Selection

•      Query curriculum DB: "Get all Organic Chemistry templates at difficulty 5-7"

•      Returns ~500 template patterns

•      Example: "Identify the structure of compound X with functional group Y"

Step 2: Parameter Permutation

•      Randomization seed ensures reproducibility

•      Randomly select 60 templates from 500

•      For each template, randomly assign parameters from available pools

•      Result: 60 unique questions

•      Probability of same 60-question paper twice: < 10^-500

Step 3: Validation

•      Verify no question repeats

•      Verify difficulty distribution (level 6 ± 0.5)

•      Verify curriculum coverage

•      Verify answer key correctness

Step 4: Encryption

•      Question paper encoded as JSON

•      Serialized to binary (protobuf format)

•      Encrypted with AES-256-GCM using Session Key

•      Add authentication tag (prevents tampering)

•      Result: Encrypted Question Blob (EQB)

•      File size: ~150KB per paper

OUTPUT: Encrypted, authenticated, difficulty-calibrated unique paper for each candidate

SECTION 3: ENCRYPTION & SECURITY PROTOCOLS

DUAL ENCRYPTION STRATEGY

Layer 1: Question Encryption (Question Blob)

Algorithm: AES-256-GCM (military-grade, unbroken since 1998)

Key: Session Key (256-bit, unique per exam session)

IV: 96 bits random per paper

Auth Tag: 128 bits (detects tampering)

Purpose: Protect questions during transmission and storage

Layer 2: Session Key Encryption (Key Blob)

Algorithm: RSA-4096 or ECDH

Key Source: Hall Supervisor's biometric hash

Purpose: Ensure only authorized Hall Supervisor can decrypt

Derivation: From secure Hardware Security Module (HSM)

RESULT: Questions encrypted with Session Key → Session Key encrypted with Biometric Hash → Only authorized person with matching biometric can access

SECTION 4: BIOMETRIC AUTHENTICATION & ACCESS CONTROL

THREE-FACTOR AUTHENTICATION

1.      Facial Recognition

–      Capture face photo at entrance

–      Compare with ID photo (Aadhaar/Passport/DL)

–      Liveness detection (anti-spoofing)

–      Match confidence: ≥99.5%

2.      Fingerprint Verification

–      Capture from 2 fingers (thumb + index)

–      Compare with Aadhaar database (via secure API)

–      Match confidence: ≥98%

3.      Voice Recognition (Backup)

–      Speak predefined phrase: "I authorize my exam to proceed"

–      Compare with voice sample from registration

–      Match confidence: ≥95%

COMBINED RESULT:

•      Probability of False Positive: 0.005 × 0.02 × 0.05 = 5 × 10^-6 (1 in 200,000)

•      Probability of False Negative: Very low (99.9% acceptance of authorized candidate)

•      System accepts authorized candidate ~99.9% of the time

•      System rejects impostor ~99.99% of the time

ACCESS CONTROL MATRIX

Role	Permission	Scope
NTA Exam Director	Initiate exam	All centers
Regional Coordinator	Monitor	10-20 regional centers
Hall Supervisor	Receive key	Single center
Invigilator	Verify identity	Single exam hall
Tech Support	Troubleshoot	Assigned center
Algorithm Engineer	Modify templates (NTA approval+2)	System-wide
Candidate	View own questions	Own paper only

 

SECTION 5: JUST-IN-TIME DEPLOYMENT WORKFLOW

TIMELINE: MINUTE-BY-MINUTE

T-2 hours: Candidates biometrically enrolled at centers

T-90 min: All centers report terminal readiness

T-60 min: Centers report "all terminals ready"

T-45 min: Biometric verification begins

T-30 min: Candidates seated; CCTV recording starts

T-15 min: All centers transmit "candidates seated" signal

T-5 min: NTA verifies "no anomalies detected" → DECISION POINT

T-2 min: EQBs + decryption keys transmitted to centers (TLS 1.3)

T-0: Keys decrypted locally; questions appear on-screen simultaneously

EXAM STARTS

After Exam:

T+180 min: Exam ends; responses locked; upload begins

T+185 min: Session keys revoked (cannot decrypt any papers)

KEY FEATURE: No question papers exist until T-2 minutes. All papers are unique to each center. Each center's paper is encrypted until T-0.

SECTION 6: DATA SEPARATION & STORAGE

THREE ISOLATED DATABASE SYSTEMS

System 1: Curriculum & Templates (NTA-Internal)

•      Contains: Templates, parameter pools, curriculum mappings

•      Compromise Impact: MEDIUM (templates exposed; requires restructuring, 2-3 weeks)

•      Encryption: AES-256 at rest

•      Distribution: 4 geographic regions

System 2: Exam Instances & Session Data (NTA-Internal, Air-Gapped)

•      Contains: Encrypted EQBs, Session Keys, center assignments

•      Compromise Impact: LOW (EQBs encrypted; keys time-expired)

•      Access: Only HSM + Key Manager

•      Storage: Air-gapped secure vault

System 3: Student Responses (NTA-Internal, Separate Server)

•      Contains: Encrypted responses, student hash, timestamps

•      Compromise Impact: LOW-MEDIUM (privacy concern; not exam integrity)

•      Encryption: AES-256

•      Retention: Per legal requirements (usually 1-7 years)

CRITICAL FEATURE: Student responses stored separately from questions. Attacker breaching response DB does not see question text.

SECTION 7: AUDIT & LOGGING

IMMUTABLE AUDIT TRAIL

Every action logged with cryptographic signature:

Log Entry Format:

{

  "timestamp": "2026-06-21T08:47:23.456Z",

  "event_type": "QUESTION_GENERATED",

  "actor": "ai_engine_neet_v2.1",

  "center_id": "NCR_001",

  "result": "SUCCESS",

  "hash_of_action": "SHA256(...)",

  "digital_signature": "RSA_SIGN(hash, NTA_private_key)"

}

Storage: Append-only database (write-once-read-many hardware)

Backup: 4 geographic regions

Signature Chain: Each log entry signs previous entry (blockchain-like)

POST-EXAM AUDIT CAPABILITY: NTA can regenerate exact same question using forensic seed → Prove authenticity

SECTION 8: CONTINGENCY & FAILOVER

HIGH AVAILABILITY ARCHITECTURE

Primary System (NTA Data Center, Delhi)

  ↓

  Backup 1 (Mumbai) - Real-time replication

  Backup 2 (Bangalore) - Real-time replication

  Backup 3 (Chennai) - Daily snapshot

RTO: <1 minute

RPO: 0 (zero data loss)

SYSTEM REDUNDANCY

Question Generation: Primary engine + pre-generated library (48-hour lookback)

Encryption Keys: HSM (primary) + Air-gapped vault (backup)

Biometric Verification: IRIS/NEC hardware (primary) + Fingerprint reader (backup) + Manual (fallback)

Network: NTA MPLS (primary) + Public internet/VPN (backup)

SECTION 9: SCALABILITY & PERFORMANCE

CAPACITY FOR NEET 2027

Peak Scenario:

•      Candidates: 24,00,000

•      Centers: 5,400

•      Sessions: 1 (OSOD - One Session, One Day)

•      Candidate per center: ~4,500

Question Generation Load:

Generate 5,400 unique papers → 2 seconds each

Sequential: 10,800 seconds (3 hours)

Parallelized (100 AI engines): 108 seconds (1.8 minutes) ✓

Network Transmission:

Transmit 5,400 EQBs (40 MB each) = 216 GB

On NTA's MPLS (100 Gbps): ~18 seconds ✓

HSM Decryption Load:

Decrypt 5,400 session keys (10ms each) = 54 seconds

With parallelization: ~5 seconds ✓

Exam Day Database:

Responses from 24 lakh candidates over 3 hours = 222 responses/second

Database throughput: 1000+ INSERTs/sec ✓

Utilization: 22% ✓

CONCLUSION: System scales easily to NEET requirements

SECTION 10: INTEGRATION WITH NTA SYSTEMS

API INTERFACES NEEDED

1.      Candidate Registration API

INPUT: Student details, Aadhaar, category

OUTPUT: Candidate ID, registration status

2.      Curriculum Mapping API

INPUT: Exam type, subject, chapter

OUTPUT: Question template IDs

3.      Center Allocation API

INPUT: Candidate ID

OUTPUT: Center ID assigned

4.      Result Upload API

INPUT: Candidate ID (hashed), responses (encrypted), timestamp

OUTPUT: Submission confirmation

EXISTING SYSTEMS (No change needed):

•      Candidate registration portal

•      Result portal (5-7 days post-exam)

•      Merit list generation

•      College allotment system (CSAB)

ENHANCED SYSTEMS:

•      Answer key publication (publish anonymized sample + unified key for study)

•      Grievance/re-evaluation (students view own paper via secure login)

•      Statistics collection (aggregate data still possible)

TECHNICAL RISK ASSESSMENT

Risk	Probability	Impact	Mitigation	Residual
Invalid question generation	LOW	HIGH	Expert validation	LOW
Network outage	MEDIUM	HIGH	Offline fallback	MEDIUM
Biometric failure	LOW	MEDIUM	Manual verification	LOW
HSM compromise	VERY LOW	CRITICAL	Air-gapped backup	VERY LOW
Algorithm leak	LOW	MEDIUM	Code signing	LOW

 

— END OF TECHNICAL SUMMARY —

This is a summary document. Full 50-page technical specification available on request including:

•      Complete pseudocode for generation algorithm

•      Detailed encryption protocols

•      Network architecture diagrams

•      Security threat models

•      Deployment procedures

•      Disaster recovery plans

Contact: hcp@RecruitGuru.com for full technical documentation