AI-powered-sql-query-assistant

The AI-Powered SQL Query Assistant is a tool that converts natural-language business questions into executable SQLite queries. It is designed for business users who may not know SQL but need to query databases. The system uses a schema-aware approach to generate SQL, validates it for read-only safety, executes it against a SQLite database, and then produces visualizations and business insights from the results. The architecture includes a Gemini 3.7 Flash model for SQL generation, a read-only SQL validator, SQLite execution, Pandas for result handling, and Plotly for visualization. Key features include explicit business metric definitions for sales, profit, orders, AOV, profit/unit, and profit margin, as well as a provider-aware LLM layer with Gemini as the current implementation. The project includes a synthetic retail analytics database with customers, products, orders, and order items, supporting realistic joins and business questions. An evaluation harness using execution/result equivalence is provided, and a frozen V4 benchmark of 10 fixed business questions achieved 90% result-equivalence accuracy and 100% SQL validation pass rate. Security measures include read-only database connections, SQL validation before execution, and blocking of destructive statements. The tool is built with Python, Gemini API, SQL, SQLite, Streamlit, Pandas, Plotly, and Git/GitHub.