How it fits together
Business question
Plain English
Schema-aware prompt
Tables and columns in context
LLM
Structured output
SQL guardrails
Read-only · validated
Reports & dashboards
<40s end to end
The challenge
Business users depended on analysts for every report. The goal was self-service analytics through natural-language questions — without the hallucinated or slow SQL that makes most text-to-SQL tools unusable on real data.
What we did
- Co-designed a hybrid AI analytics platform with workspace-based dashboards and automated reports.
- Built a schema-aware natural-language-to-SQL pipeline using structured outputs.
- Added safety guardrails so generated SQL is read-only, validated, and performant across arbitrary datasets.
- Optimised prompts and query plans to cut end-to-end latency.