A design surface, not a black box.
LucidOntos exposes every layer you'd build by hand — ontology graph, mapping rules, dialect codegen, pipeline shape, reconciliation logic — so you stay in control while skipping the boilerplate.
The pain today
- Hand-writing 80% boilerplate SQL for every new mart.
- Refactoring Bronze/Silver/Gold every time the source schema shifts.
- Re-implementing SCD2, dedupe, PII masking from scratch per project.
- Reconciliation is the script no one wants to own.
- Changing a target dialect (Snowflake → Databricks) is a quarter-long rewrite.
What LucidOntos delivers
- Portable ontology: target-neutral entities, relationships, SCD strategy, history flags.
- Auto-mapping with explicit transforms — CAST, TRIM, COALESCE, env-conditional SHA2.
- Native dialect codegen: Snowflake SQL, Snowpark, PySpark, Delta, BQX, Glue.
- Pipeline shape you can read: extract → bronze → silver → gold → reconcile.
- Reconciliation rules emitted with the DAG — row counts, checksums, key aggregates.
- Re-run codegen on a new target: same ontology, fresh dialect-native code.
Artifacts you'll review
Ontology graph
Customer → Account → Order → Invoice → Payment, with SCD2 flags and PII tags.
Mapping table
Source column → target column → transform → confidence → review status.
Generated SQL / PySpark
Side-by-side native code for your chosen platform. Edit, regenerate, version-control.
DAG / pipeline
Airflow, ADF, Composer, Databricks Workflows, Fabric Pipelines — emitted from the same ontology.
Reconciliation rules
Per-batch row count parity, checksums, business-key aggregates — wired into the DAG.
AI Health Analysis
When a Gold mart fails, get the failing column, the root cause, and a proposed fix.
A real failure from the demo project: Gold broke because currency_code was null on 412 invoices from 2024-03-12. The AI Health Analysis named the column, the date range, and proposed the COALESCE fix — in clear, human-readable terms. That's the design surface working with you, not for you.