​
AI-Powered Assessment

When a review task is created, Velatir’s AI assessment engine evaluates it against all active policies using advanced language models.

​
Assessment Flow

1

Request Analysis

AI examines function name, arguments, explanation, and metadata
2

Policy Application

Each active policy is applied to the request
3

Risk Evaluation

AI determines compliance status and risk level
4

Recommendation Generation

System recommends auto-approval or human review

​
Assessment Criteria

​
Function Analysis

  • Function Name: Patterns indicating sensitive operations
  • Arguments: Data types and content analysis
  • AI Explanation: Understanding of the AI’s reasoning
  • Metadata: Additional context and session information

​
Policy Evaluation

For each policy, the AI evaluates:
  • Compliance: Does the request meet policy requirements?
  • Risk Level: What’s the potential impact? (Low/Medium/High/Critical)
  • Confidence: How certain is the assessment? (0.0-1.0)
  • Tags: What categories apply? (#PersonalData, #FinancialData, etc.)

​
Security Measures

  • Prompt Injection Protection: Ignores instructions within request data
  • Consistent Standards: Maintains evaluation criteria across requests
  • Conservative Approach: Errs on side of requiring review when uncertain