Skip to content
WIZZARD.WORK
Work index
03.04Case study

dremio-ai-agents

Agents that query a Dremio lakehouse and defend the SQL they wrote.

Two translucent grid printed sheets overlapping on white paper, with a navy line across the middle.
Role
Author
Stack
  • Dremio
  • Python
  • MCP
  • SQL
View the repository
Last push
1y ago
03.04.1

Problem

Text-to-SQL against a real semantic layer fails quietly. The query runs, returns rows, and the rows are wrong.

03.04.2

Approach

The agent reads the semantic layer first, drafts SQL, then runs a validation pass against row counts and known aggregates before returning anything.

03.04.3

Outcome

Answers arrive with the query, the tables touched, and the checks that passed. An analyst can audit the path in under a minute.