Platform Architecture
Connect
Read-only metadata extraction via native connectors
Extract
Schema structure & small data samples (5 rows)
AI Process
Multi-model ensemble analysis & enrichment
Export
JSON, HTML, Markdown, or PDF documentation
Read-Only Extraction
We extract schema metadata and small samples (5 rows) only. No bulk data movement.
In-Memory Processing
All AI analysis happens in-memory. Data is not persisted after the session.
AI Description Generation
3X uses an ensemble of three advanced AI models (Claude, GPT-4, and our fine-tuned custom model) to analyze your data from multiple angles.
Naming Conventions
Translates technical column names into human-readable labels (e.g., cust_id → Customer ID).
Data Patterns
Infers semantic meaning from actual data formats (regex matching & sampling).
Relationships
Analyzes foreign keys to understand how business entities relate to each other.
Confidence Scoring
Trust but Verify
Every generated description gets a 0-100% confidence score. This prioritization engine helps you focus your manual review where it matters most.
| Score Range | Meaning | Recommendation |
|---|---|---|
| 90-100% | High Confidence | Minimal review needed. Likely accurate. |
| 70-89% | Medium Confidence | Quick verification recommended. |
| < 70% | Low Confidence | Human review strongly recommended. |
Intelligent PII Detection
Our multi-layered approach combines pattern matching, semantic analysis, and data sampling to detect sensitive data automatically.
Contact Info
Emails, Phone Numbers, Physical Addresses
Personal IDs
SSNs, National IDs, Passport Numbers
Financial
Credit Card Numbers, Bank Account Details
Health
Medical Record Numbers, Health IDs
Human-in-the-Loop Workflow
The Philosophy
AI handles the 80% of straightforward documentation. Humans focus their expertise on the 20% requiring complex business context.