Join GDG Hong Kong for an intensive session focused on Web AI Agents and Edge Computing. We are honored to host Jason Mayes, Google's Web AI Lead, along with local GDEs. We will explore how to build AI applications that are fully offline, privacy-first, and completely free of hosting costs using LiteRT.js and the new Gemma 4 models. From revolutionizing special education to building autonomous web assistants, this event showcases the real-world impact of AI in the browser.

Agenda & Talk Details:

1. Client-side Web AI Agents for the Agentic Internet of the Future

• Speaker: Jason Mayes (Web AI Lead, Google)

• Synopsis: AI usage in JavaScript is growing exponentially. Jason will demonstrate how generative models from Google running entirely on the front end can perform agentic behaviors—giving your websites "superpowers" to automate tasks.

  • Key Takeaways: Privacy, low latency, and zero inference costs.

  • Language: English | Level: Intermediate | Duration: 30 mins

2. WebAI Empowered Oculomotor Rehab for SEN (「視」界無界：基於WebAI的兒童眼動復康與邊緣大模型輔助平台)

• Speakers:

  • Cyrus Wong (黃俊彥) – Senior Lecturer, Hong Kong Institute of Information Technology (HKIIT) at IVE (Lee Wai Lee) & Google Developer Expert - Cloud(AI)

  • Esther Cheung (張曉彤) – Higher Diploma in Cloud and Data Centre Administration (雲端系統及數據中心管理高級文憑) - Hong Kong Institute of Information Technology (HKIIT) at IVE(Lee Wai Lee) & GDG on Campus - HKIIT Lead

• Synopsis: Discover how WebAI is breaking financial barriers in special education. Learn how a revolutionary low-cost ecosystem uses standard webcams to provide precise eye-tracking for children with Special Educational Needs (SEN).

  • Tech Stack: MediaPipe Iris, WebGazer.js, LiteRT.js, and Gemma.

  • Key Takeaways: High-precision gaze tracking without server costs; 100% data privacy for sensitive users.

  • Language: Cantonese | Level: Intermediate | Duration: 30 mins

3. Private and Offline: On-Device AI with LiteRT.js and Gemma 4

• Speaker: Connie Leung - GDE (Angular, AI, Web)

• Synopsis: Get hands-on with LiteRT.js and Google’s open-source Gemma 4 models. Learn to retrieve models from Hugging Face and leverage WebGPU for high-performance in-browser inference.

  • Key Takeaways: Setting up LiteRT-LM; optimizing WebGPU for browser acceleration; designing offline-first security.

  • Language: English | Level: Intermediate | Duration: 30 mins

4. Airplane Mode Agents: SKILL.md on a Phone and Browser

• Speaker: Sasha Denisov - CTO & Co-founder at Brainform.ai · Google Developer Expert (Flutter, Firebase, Cloud AI) & Maintainer of flutter_gemma

• Synopsis: Running an AI model locally is nearly solved—determining what that model is allowed to do is the next major hurdle. On-device agent skills require three core elements: a portable description, discovery, and invocation. While descriptions and discovery have converged around markdown definitions and tool lists, invocation remains fragmented across local code, native OS intents, and remote tool calls (like MCP servers). In this lightning session, Sasha introduces the unified skill layer built on top of flutter_gemma, demonstrating how plain SKILL.md files provide a single dispatch path for local, native, and remote agent actions—even when completely offline in "Airplane Mode".

  • Key Takeaways: Unified On-Device Execution - How to use portable SKILL.md files with flutter_gemma and LiteRT-LM to route local code, OS intents, and remote tool calls through a single, offline-first dispatch path.

  • Language: English | Level: Intermediate | Duration: 10 mins (Lightning Talk)

Target Audience / Prerequisites:

• Web developers (JavaScript/TypeScriptFlutter/Dart), AI enthusiasts, and tech professionals.

• Familiarity with web fundamentals (HTML/CSS/JS) is recommended.

• Please bring your own laptop to follow the live demos and code snippets.
  

Event Details:

• Date: 19 August, 2026

• Time: 7:30 PM – 9:30 PM HKT (Check-in starts at 7:00 PM)

• Venue: Google Hong Kong, 21/F, Tower 2, Times Square, Causeway Bay.

• Light refreshments and networking session included.

=== Special thanks to our venue partner and refreshment sponsor Google Hong Kong ===