Skip to main content

Automatic Variation Engine

Overview​

SuiVerse's core technology, the automatic variation engine, transforms original quizzes in real-time to provide unique questions to each user. This prevents question leaks at the source and ensures fair assessment.

Variation Generation Flow​

Variation Techniques​

1. Synonym Substitution​

Replace original words with other words of the same meaning.

Original: "What are the capabilities of struct in Sui Move?"
Variation: "What are the functions of struct in Sui Move?"

2. Structure Transformation​

Change the structure of sentences while maintaining meaning.

Original: "Select the correct answer from the following"
Variation: "Choose the correct answer"

3. Voice Change​

Change the voice of sentences, including active/passive transformations.

Original: "The developer deploys the smart contract"
Variation: "The smart contract is deployed by the developer"

4. Hypernym Replacement​

Replace specific terms with more general terms or vice versa.

Original: "Issue an NFT"
Variation: "Create a digital asset"

5. Number Formatting​

Change the representation format of numbers.

Original: "100 questions"
Variation: "one hundred questions"

6. Formality Adjustment​

Adjust the formality level of sentences.

Original: "What is the correct answer?"
Variation: "What would be the correct answer?"

Quality Assurance​

Variation Quality Verification​

  • Meaning Preservation Check: Verify that original and variation have the same meaning
  • Grammar Check: Validate grammatical accuracy of transformed sentences
  • Difficulty Consistency: Maintain difficulty level after variation
  • Readability Check: Confirm clarity and comprehensibility of transformed questions

Quality Score Criteria​

ScoreStatusAction
90-100%ExcellentUse immediately
70-89%GoodAvailable for use
50-69%Needs re-variationApply different variation technique
Below 50%FailedUse original or manual processing

Security Features​

Zero-Knowledge Principles​

  1. In-memory Decryption: Original is temporarily decrypted only in memory
  2. Immediate Deletion: Immediately deleted from memory after variation completion
  3. No Logging: Original content is not logged anywhere
  4. No Access: Even administrators cannot access original content

Unique Seed Generation​

function generateSeed(userId: string, examId: string, timestamp: number): string {
    return hash(userId + examId + timestamp + randomNonce());
}

Performance Optimization​

Caching Strategy​

  • User-specific Caching: Use cache for repeat requests from same user
  • TTL Setting: Auto-delete after 24 hours
  • Memory Management: Ensure memory efficiency with LRU cache

Parallel Processing​

  • Batch Variation: Transform multiple questions simultaneously
  • Asynchronous Processing: Handle variation tasks asynchronously
  • Load Balancing: Distribute work across multiple workers

Statistics and Monitoring​

Variation Statistics​

  • Daily variation count
  • Average variation time
  • Quality score distribution
  • Cache hit rate

Performance Metrics​

  • Variation latency: Below 100ms
  • Throughput: 1000 variations/second
  • Memory usage: Below 2GB
  • CPU usage: Below 70%

Future Improvement Plans​

  1. AI Model Enhancement: Generate more natural variations
  2. Multi-language Support: English, Chinese, Japanese, etc.
  3. Context Awareness: Consider context of entire question sets
  4. Adaptive Variation: Adjust variation intensity to user level