I completed a toy setup of GEPA with some fricken good results! I am very happy to have a methodology that removes some of the guess work with prompt design. Its a really simple algorithm that orchestrates a genetic evolution where the mutation operator is asking a LLM to improve the prompt based on the diagnostic trace your evals. So simple! I also enabled websearch during reflection so it can actually do its research when improving the prompt. This means it would adapt the web documentation to suit the prompt format automatically. Very good, been looking for something like this and it did not disappoint.
đź“ť Trying out "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning"
https://arxiv.org/abs/2507.19457 25 Jul 2025 Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong, Arnav Singhvi, Herumb Shandilya, Michael J Ryan, Meng Jiang, Christopher Potts, Koushik Sen, Alexandros G. Dimakis, Ion Stoica, Dan Klein, Matei Zaharia, Omar Khattab -- UC Berkeley, Stanford University, http://BespokeLabs.ai|BespokeLabs.ai, Databricks, MIT Replicating GEPA LLMs are powerful, but performance hinges on the prompt. I wanted to explore GEPA (Genetic-Pareto) as a method for optimizi
RIP “Prompt Engineer” 2022-2025 the latest human job to fall to AI