BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.coscup.org//coscup-2026//talk//QNFNF7
BEGIN:VTIMEZONE
TZID:CST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:CST
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-coscup-2026-QNFNF7@pretalx.coscup.org
DTSTART;TZID=CST:20260809T130000
DTEND;TZID=CST:20260809T140000
DESCRIPTION:Everyone is saying "AI" right now. Fewer people are saying whet
 her it actually made their work better. I build this stuff for a living 
 — an on-device AI appliance around a Raspberry Pi 5 + AI HAT+\, and mobi
 le apps running speech and language models locally: sherpa-onnx\, whisper.
 cpp\, llama.cpp\, Gemma\, Style-Bert-VITS2. Not in the cloud\, on the devi
 ce\, for people whose data shouldn't leave it. Mostly what I've learned is
  how it breaks. Small models forget what you just told them. Latency you c
 an feel. Demos that impress someone once and never get opened again. So th
 is is not a talk with answers. It's a room for people actually shipping th
 is stuff to compare notes and be honest:\n Where has local AI genuinely ma
 de something better — and where did we just bolt a chatbot onto a thing 
 that was fine?\n \nWhat's still missing? Models\, tooling\, hardware\, UX 
 patterns. What would you build if someone else built the piece you keep re
 writing? \n\n\nReal constraints: what did you give up to run on-device\, a
 nd what turned out not to matter? \nWe'll run it Lean Coffee style — eve
 ryone writes down the question they came with\, we vote\, and we work down
  the list. No conclusions required. "It'd be nice if this existed" is a pe
 rfectly good outcome. A Raspberry Pi 5 will be running fully offline on th
 e table the whole hour\, so when the conversation drifts too abstract we c
 an go poke at something real. Anyone welcome: edge AI\, on-device LLM / AS
 R / TTS\, low-resource languages\, hardware — or you just have a strong 
 opinion about whether any of this is working yet. Newcomers especially wel
 come. The questions are the point. \n\n大家都在講 AI，但很少人
 講它到底有沒有讓工作變好。\n我平常做的是端上的 AI：
  Raspberry Pi 5 + AI HAT+ 的裝置，以及把語音與語言模型跑在
 手機本機的 app （sherpa-onnx、whisper.cpp、llama.cpp、Gemma、Sty
 le-Bert-VITS2）。做下來學到最多的 其實是它怎麼壞——小
 模型記不住剛講過的話、延遲感覺得出來、demo 驚艷一次
 之後再也 沒人打開。 這場不是有結論的演講，而是想找
 同樣在做的人誠實聊聊：哪些地方地端 AI 真的讓事情 變
 好了？哪些只是硬加了一個 chatbot？還缺什麼模型、工具
 、硬體或 UX pattern？我們 用 Lean Coffee 的方式進行，各自
 把帶來的問題寫下來、投票、依序討論。不需要結論， 
 「有這個就好了」也是很好的收穫。桌上會一直放一台
 完全離線運作的 Raspberry Pi 5， 聊到太抽象的時候可以回
 來戳戳實體。歡迎做邊緣 AI、端上 LLM/ASR/TTS、低資源語
 言、 硬體的朋友，或單純對「這些到底有沒有用」有意
 見的人。新手非常歡迎。 Language: English (Mandarin / Japanese a
 lso fine)
DTSTAMP:20261022T191658Z
LOCATION:TR310-2
SUMMARY:[BoF] What Is Local AI Actually For? - 
URL:https://pretalx.coscup.org/coscup-2026/talk/QNFNF7/
END:VEVENT
END:VCALENDAR