BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.coscup.org//coscup-2026//speaker//38ZMFE
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-2026-87EJBH@pretalx.coscup.org
DTSTART;TZID=CST:20260809T100000
DTEND;TZID=CST:20260809T103000
DESCRIPTION:flutter_gemma started as a thin wrapper over MediaPipe for runn
 ing Google Gemma on Android and iOS. Today it's a full-fledged platform fo
 r on-device AI: 6 platforms (Android\, iOS\, Web\, macOS\, Windows\, Linux
 )\, 2 inference engines\, multimodal support\, function calling for local 
 agents\, and on-device RAG with vector search. The plugin has been feature
 d by Google AI for Developers.\nIn this talk\, we'll walk through the evol
 ution via key engineering decisions. How adding Web platform forced us to 
 rethink file handling and led to sealed classes instead of string URLs. Wh
 y we had to spin up a gRPC server in Kotlin with a bundled JVM for desktop
  — and how to automate the build through Xcode build phases. How the Str
 ategy pattern allowed us to add a second inference engine (LiteRT-LM) with
 out rewriting existing code\, and how the Adapter pattern helped reuse the
  MediaPipe implementation. Why we needed chunk buffering for engines with 
 fundamentally different APIs.\nWe'll dive deep into on-device RAG: how to 
 build a SQLite VectorStore that works identically on mobile and in the bro
 wser via WASM. I'll show real bugs and their fixes — from Web hot restar
 t crashes to iOS Simulator limitations with vision models.\nThe latest add
 ition is genkit_flutter_gemma — a bridge to Google's Genkit for Dart tha
 t enables hybrid AI pipelines. Now you can seamlessly combine on-device in
 ference with cloud-based models in a single Genkit flow: run lightweight t
 asks locally for speed and privacy\, and escalate complex reasoning to the
  cloud — all orchestrated through one unified pipeline.\nEvery architect
 ural decision in flutter_gemma is an answer to a specific problem. Minimum
  theory\, maximum code\, diagrams\, and stories of "how it broke and why i
 t looks like this now".
DTSTAMP:20260601T180905Z
LOCATION:RB102
SUMMARY:Evolution of flutter_gemma: From a Simple Wrapper to a Cross-Platfo
 rm AI Engine - Sasha Denisov
URL:https://pretalx.coscup.org/coscup-2026/talk/87EJBH/
END:VEVENT
END:VCALENDAR
