Normalise. Prep.
Govern.

The step data engineers skip. Xynage detects type mismatches across columns, composite primary keys, orphaned columns with no downstream consumers, and naming convention conflicts. The agent proposes a normalised DDL. Your engineer reviews and approves before any data moves.

LLM-assisted cleaning rules — format standardisation, null strategies, value normalisation (country codes, phone numbers, date formats), PII detection and masking before any LLM call. Every cleaning rule is proposed by the agent and approved by your team. Cleaning log is committed to git.

The agent generates a dbt semantic model from the clean prep tables — dimension hierarchies, measure definitions, entity relationships. You approve the model before it deploys. The result is a reproducible mart layer that downstream BI tools and ML pipelines can consume directly.

Probabilistic entity resolution using Splink with configurable blocking rules and EM training. The agent surfaces duplicate candidate pairs ranked by match score. Stewards review and approve or reject each merge decision. Every decision is logged to the approver. Golden records are versioned — merge history preserved.