How I Built a Smart SQL Optimizer with AI

Summary

The presenter demonstrates a Python-based SQL optimizer built with Microsoft AutoGen. It is designed to help developers who are less experienced with optimizing Oracle Database queries.

Two-agent workflow

• A SQL analysis agent examines the query, retrieves its execution plan, checks existing table indexes, and gathers index-cost information.
• An email agent sends the results.

AI recommendations

The analysis agent uses GPT-4o in the demo, though the presenter notes that another LLM could be substituted. Recommendations may include creating an index when a query performs a full table scan, along with an updated query.

Oracle Cloud deployment

The AutoGen code is deployed as an OCI Function. A thread-pool executor runs the asynchronous agent code, and the presenter demonstrates calling the function with a SQL query payload.

Results and operations

The example completes in 22 seconds and reports that the email was sent. OCI Logging is used to inspect function logs and assist with debugging. The emailed results include optimization advice. The presenter cites potential performance improvements of 20–30% for similar queries in certain scenarios, not as a universal guarantee.

Integration options

The function can be invoked directly or connected to event-driven workflows, such as a queue or stream.

Guide or tutorial covered

The video is a code walkthrough and deployment demo. It explains the agent roles, SQL analysis steps, email delivery, OCI Function execution, and log inspection. It does not provide a detailed, line-by-line code tutorial.

Main speaker/source: The presenter, attributed to Christophe Pruvost.