the find
− 0.6 mA native iPhone app for Thai speakers who are deaf or hard of hearing, living abroad in an English-speaking world. Point the phone at a conversation and it produces live Thai captions — separated by speaker, so a group conversation reads like a script rather than a wall of text. The name is the promise: ฟังได้, "can hear."
It sits at the intersection of several strata at once — Thai/English bilingualism, years of speech-AI work, and native app craft — aimed at the one audience that benefits most from all three.
cross-section
− 1.3 mBy default everything runs on the phone. On-device speech recognition feeds Apple's on-device translation; speaker diarization (via FluidAudio) splits the transcript by voice before translation, so each caption carries its speaker. An optional revision pass cleans the Thai after the fact — on the phone with Apple's own model, or through a cloud model only if the user adds a key. No audio leaves the device — a conversation is the most private data there is, and the architecture treats it that way. Swift, generated project config, App Store listing drafted, formal evaluation run 2026-07-12.
inscription
− 2.0 mField note, from the design review: The hard problem was never translation — it was who said what. Captions without speakers are noise in any language. Running diarization on-phone, in real time, ahead of translation, is the feature that turns a transcript into a conversation you can follow.