Spent months embedded at Cursor building the distributed inference infrastructure for Composer 2's large-scale RL training, including global cluster orchestration, custom kernels, and weight synchronization systems.
no scored calls yet — needs a stated position or a categorical verdict, with a matured window vs SPY
Fireworks provides the inference infrastructure layer that makes large-scale RL economically viable by disaggregating training and inference across global GPU clusters, using custom kernels (FP4, router replay for MoE) and delta weight compression to synchronize 1TB model snapshots across continents in under a minute.