We may have had many sparks of performance optimization, and wanna a quick PoC to verify if one idea works or not. But this procedure is very time-consuming, especially in a mature and complex system. This topic shares my practice of leveraging LLM Agent to accelerate my performance optimization tasks in real product systems by having agents do “quick tries” for me and leave the precious human decision to real critical things. And some lessons learned from it, like this method can’t reduce your work but makes you even more tired :P

Speakers:

Ruihang Xia: Software Architect at Greptime Inc.

Software Architect at Greptime Apache DataFusion PMC, Apache Arrow Committer, Apache HoraeDB PPMC https://github.com/waynexia