From Manual Prompt Tuning to AI Tuning: Genetic Algorithm-Based Automated Optimization and Acceleration

LY Corporation’s Tech-Verse 2026 article explains how Yahoo! JAPAN Search automates LLM prompt tuning with GEPA, a genetic-algorithm-based optimizer.

The system iteratively improves candidate prompts using natural-language reflections and multi-criteria evaluation.

Implemented with DSPy, the evaluation function returns both a scalar score and feedback to guide optimization.

The approach cuts manual prompt-adjustment time dramatically, from days or weeks to about one hour, even in fast-changing model environments.