Build and own a Semantic Mapping Specialist Model.
Mapping business-question terms and entities (e.g. 'revenue', 'Northeast region', 'churn') onto the actual schema objects (tables, columns, metric definitions) in a company's data warehouse or semantic layer.
Overview
Reduces incorrect query results caused by mapping a business term to the wrong table or column.
Who owns this decision
VP Analytics, Chief Data Officer
What this Specialist Model actually does
Decision: Mapping business-question terms and entities (e.g. 'revenue', 'Northeast region', 'churn') onto the actual schema objects (tables, columns, metric definitions) in a company's data warehouse or semantic layer.
This is a larger system with 1 specialist model candidate and 4 supporting or external components.
Who uses this Specialist Model
- VP Analytics
- Chief Data Officer
- Head of BI
- VP Data Platform
Where the specialist model sits
This is a larger system with 1 specialist model candidate and 4 supporting or external components.
Ambiguity detector routing unclear mappings
Specialist Model
Ambiguity detector routing unclear mappings to human confirmation
Structured extraction/retrieval model grounding business terms in a live schema index
Supporting component
Semantic-layer or metrics-layer integration where one already exists
Supporting component
Read the Semantic Layer guide →
Historical query-log retrieval
Supporting component
Historical query-log retrieval for term-to-schema precedent
Schema-change monitor
Supporting component
Schema-change monitor to invalidate and refresh mappings when the warehouse evolves
How Oumi connects to your stack →
How to evaluate and improve it
Evaluation
- Mapping accuracy vs. human-labeled ground truth
- Ambiguity-detection precision and recall
- Consistency of mapping the same term across repeated requests
Feedback signals
- Analyst confirms, edits, or rejects a proposed schema mapping
- Generated SQL query fails or returns implausible results traced to a wrong mapping
- Same business term mapped differently across separate requests (inconsistency signal)
- Human correction of a mapping during data-team QA review
- New schema/table added that changes the correct mapping for an existing term
Model candidates to evaluate
- NuExtract 3
- Qwen3.5-9B
These are candidates, not performance claims. Run them against your evaluation set.


