Portfolio
QueryPilot: Natural Language to SQL Automation Platform
Converts plain English questions into safe, accurate SQL using schema ingestion, constrained prompting, validation, and automatic error-aware correction.
Project details
What we delivered
Overview
Natural language interface for analytics with schema grounding.
Produces read-only queries with automated correction.
Why this exists
SQL bottlenecks block non-technical teams from data.
Text-to-SQL fails mainly on schema linking and correctness.
How it works
Schema discovery and relationship mapping.
Schema-grounded SQL generation.
Execute with SELECT-only and least privilege.
Error-aware retry loop on database errors.
Key engineering decisions
Constrained output to reduce invalid SQL.
Read-only enforcement and allowlisting.
Reliability via auto-repair loop.
Optional local execution for privacy.
Use cases
Self-serve analytics.
Automated KPI reporting.
Support insights.
Operations and finance queries.
Tech stack
Schema ingestion with context selection.
LLM prompting with strict formatting.
Database execution sandbox with read-only access.
Optional orchestration integration.
What MuFaw delivered
Schema ingestion and relationship mapping service.
Guardrailed SQL generation with repair loop.
Read-only execution layer and integration hooks.