Kairo

Kairo is a research workbench and experimental system for optimized workload measurement and inference routing on NVIDIA RTX 5090 GPUs. It features evidence-driven Blackwell inference research, defining versioned workload protocols, declarative kernel blueprints, and selective runtime policy based on exact measured workloads. Kairo is designed for researchers and developers focused on GPU inference optimizations but is not a general-purpose inference engine.