OpenRSI × SGLang-Omni: Scaling RSI for AI Infra

OpenRSI and the SGLang-Omni team report that applying recursive self-improvement to AI infrastructure through automated agents and independent verification raised omni model throughput on H100 by 7.6% to 326.5%. The agents operate in a continuous loop to propose and test infrastructure optimizations, while human experts define the tasks, handle system design, and review the resulting pull requests. Open article →