Episode Details

A listener running a Futo voice input fork on Android wants to chain a custom small model after his ASR and punctuation stages to fix number normalization, punctuation, and paragraph breaks. But number normalization, punctuation restoration, and paragraph segmentation aren't one task — they're three tasks with different shapes, different data requirements, and different latency costs. This episode works through the whole design space: multi-stage pipelines versus single purpose-built models like Superwhisper's S1, why Replicate's Cog format fights against four-model chains, what actually changes when everything has to run on a phone, and how many training samples each transform really needs. Episode #067442 — open it directly at myweirdprompts.com/067442