Ask Your Database in Plain English: Natural Language SQL

The Data Access Problem

Only 3 people on your team can write SQL. Everyone else files a ticket and waits 2 days. NL-SQL changes this — type a question, get data instantly.

How It Works

1. User asks: "Top 10 customers by revenue last quarter?"
2. LLM receives question + schema, generates SQL
3. System executes on a read-only replica

async function nlToSql(question: string, schema: string) {
  const res = await openai.chat.completions.create({
    model: 'gpt-4o',
    messages: [
      { role: 'system', content: `SQL expert. Schema:\n${schema}\nConvert to SELECT only.` },
      { role: 'user', content: question }
    ],
    temperature: 0,
  });
  return res.choices[0].message.content;
}

Safety Guardrails

• Read-only — No writes or deletes
• Timeout — Kill after 10 seconds
• Row limit — Cap at 1,000 rows
• Schema filtering — Hide sensitive tables

Accuracy

Query Type	Accuracy
Simple aggregation	~95%
Multi-table join	~85%
Time-series	~80%
Complex subqueries	~65%

Use NL-SQL for ad-hoc exploration, dashboards for recurring metrics.