SQL for GenAI

Beginner → Intermediate · 6 topics

Most company data lives in SQL databases — orders, customers, tickets, payments. GenAI engineers meet SQL in three places:

1. Feeding LLM apps: fetching the customer's orders for a support bot, filtering documents for RAG.
2. Measuring LLM apps: LLM call logs, costs, latencies and eval results are tables. SQL answers "which feature costs the most?" or "did prompt v3 beat v2?" in seconds.
3. Text-to-SQL: letting users ask questions in plain English and having an LLM write the query — one of the most requested enterprise GenAI features, and one of the riskiest.

All examples use SQLite, which comes with Python — nothing to install, and every query runs. The SQL is standard and works in PostgreSQL and MySQL with small changes (noted where they matter).

import sqlite3                     # built into Python — no pip install needed

#	Topic	Sub-topics
1	SQL basics with Python	Tables · SELECT · WHERE · ORDER BY · LIMIT · sqlite3 · parameterised queries
2	Joins & aggregation	GROUP BY · HAVING · INNER / LEFT JOIN · subqueries · NULL traps
3	Advanced SQL	CTEs · window functions · CASE · dates · JSON columns · indexes · pgvector
4	Querying LLM logs & evals	Cost per feature · p95 latency · error rates · comparing prompt versions · regressions
5	Text-to-SQL	Schema in the prompt · few-shot · validating SQL · read-only safety · evaluating accuracy
6	Text-to-SQL agents	Tools for exploring a database · the agent loop · self-correction · production guardrails

Where this fits

Read after Python Basics. Pandas comes next — and pd.read_sql() turns any query from this series into a DataFrame.

Next: Pandas — tables of data in Python.