Skip to main content

​
The Fabricator

The Fabricator is Flightline’s synthetic data engine. It transforms the scenarios from the Cartographer into high-fidelity, internally-consistent test data.

​
The Problem It Solves

Developers at companies handling sensitive data face a blocking problem:
“I need to test my LLM feature against realistic credit reports, tax documents, and loan applications. But I can’t download production data due to PII and security policies.”
The Fabricator generates production-grade synthetic data that:
  • Matches your exact schema structure
  • Follows all business rules and constraints
  • Contains zero PII, making it legally safe to use anywhere

​
How It Works

Input:  Scenario from Cartographer
        "High credit, low income, Jumbo loan"
        ↓
Process: Apply schema constraints
         Apply business rules
         Generate consistent values
        ↓
Output: Valid JSON record

​
Schema Parsing

The Fabricator understands your schema structure, including field types, constraints, and relationships.

​
Business Rule Enforcement

The Fabricator infers and enforces business rules:
  • If loan_type = "Jumbo", loan amount exceeds conforming limits
  • If employment_status = "Employed", income is greater than zero
  • If bankruptcy = true, bankruptcy date is in the past

​
PII Synthesis

PII fields are detected and replaced with realistic fakes:
Field TypeExample Output
Name”Jordan Smith”
Email”j.smith.482@example.com”
SSN”482-39-1847” (fake, valid format)
Phone”(555) 482-3918”
Generated identifiers follow valid formats but are not real. They’ll pass format validation but aren’t tied to real people.

​
Internal Consistency

The key differentiator is internal consistency. Random data generators produce nonsensical combinations: 
// Random generation (inconsistent)
{
  "employment_status": "Employed",
  "income": 0,
  "employer_name": null
}

// Fabricator output (consistent)
{
  "employment_status": "Employed", 
  "income": 85000,
  "employer_name": "Acme Corp"
}

​
What’s Next

Once data is generated, the Fact-Checker validates your LLM outputs against it.

The Cartographer

Learn how scenarios are mapped before generation.

The Fact-Checker

Validate LLM outputs against generated data.