NL2SQL Core Engine — 6 Pillars

Prevent NL2SQL regressions when switching LLM providers by running a 23-case automated evaluation suite to generate test_report.json, verifying the multi-agent pipeline maintains 23/23 PASS before deploying changes.

Schema linking improves NL2SQL query generation by mapping user intent to correct tables and columns while injecting value profiles, ensuring generated SQL uses exact enum values and reducing invalid query execution.

Fix SQL dialect errors in an NL2SQL pipeline by applying centralized repair patches and regex modifications for specific date phrases, enabling the system to self-recover from dialect-breaking syntax issues automatically.

Automate NL2SQL prompt tuning and security validation together by configuring Gatekeeper validation rules with injection test cases like DROP TABLE, automatically classifying valid analytical questions versus unsafe requests.

NL2SQL ambiguity detection is the process of tuning Gatekeeper rules to classify valid analytical questions versus unsafe or invalid requests, needed when extending synonyms or adding schema updates to maintain pipeline safety.

Your NL2SQL pipeline fails after adding new tables because schema linking, value profiles, and prompt schema text require manual updates, which must be verified by running test_pipeline.py to achieve 23/23 PASS.