Projects 2026

Q-Cards

Fine-tuning LLMs for database query optimization

I designed and implemented a new fine-tuning pipeline that trains large language models to produce hints for a database’s query optimizer.

Results

On queries the model had never seen:

  • Execution latency dropped by 26%, and by up to 91% on complex queries.
  • Energy use dropped by 23%.
  • It outperformed prior work on high-complexity queries.