DevFest Montréal 2026

← Back to the programme
4-5 · Never Gonna Give You UpEnglish

Flight Mode AI: Shipping Reliable Offline AI with Flutter and Dart

Category
AI & ML
Format
Session — 40 minutes
Audience level
Intermediate

Cloud AI is powerful until the network disappears. I learned this while building QuAI, a writing assistant I use every day. The interface could stay the same, but a reliable offline route required much more than running llama.cpp on my laptop. In this session, we’ll turn that prototype into a cross-platform product. We’ll trace how llamadart delivers prebuilt runtimes at build time and exposes llama.cpp and LiteRT-LM through one Dart API across mobile, desktop, and web. Then we’ll use ordinary Flutter and Dart code to stream local chat, reuse cached models, and switch a Genkit app between Gemini and an in-process local model. Expect architecture, practical trade-offs, and live demos, not a tour of C++ build flags. You’ll leave knowing how to add a dependable offline path to an AI app, what each runtime can support, and why offline AI is a product reliability and privacy feature, not just a fallback for airplane mode.

Speakers

Jhin Lee

Senior Software Developer · Unity

Jhin Lee is a Senior Full-Stack Developer at Unity and a Google Developer Expert (GDE) in Flutter and Dart. With over 20 years of hands-on experience across embedded systems, backend infrastructure, Android, iOS, and web, Jhin focuses on building robust solutions that push the boundaries of modern architecture. Driven by a passion for open-source tooling and AI integration, Jhin is the author of numerous projects, including llamadart (a package for cross-platform local LLM inference via llama.cpp), dinja, and mcp_dart. Beyond daily development, Jhin is a community leader serving as an organizer for GDG and Flutter Montreal, frequently speaking on seamlessly integrating advanced AI models and the realities of maintaining open-source libraries.