ChaGather: Mindful AI Tea Companion

Find inner peace at home. Co-designed with the founder of Ottawa's ChaDynasty tea house, Master Chady uses Gemini Live to guide authentic, calming Gongfu tea ceremonies in your living room.

Comment

Inspiration

In our fast-paced modern world, finding true inner peace is harder than ever. Traditional Chinese Gongfu tea ceremonies offer a profound, meditative escape, but the barrier to entry is high—most people don't know the exact temperatures, steeping times, or the mindful flow required to host one at home.

We wanted to bridge that gap using ambient AI. We collaborated directly with the founder and lead tea master of ChaDynasty, an authentic tea house based in Ottawa, to digitize the experience of a guided tea ceremony. By combining their deep, real-world domain expertise with the real-time, multimodal streaming capabilities of the Gemini Live API, we created a digital tea companion that brings the serenity of a physical tea house right to your living room table.

What it does

Master Chady (via the ChaGather app) is a voice-activated, camera-aware digital tea pet. Instead of a traditional chat interface, it uses an "Invisible UI"—a calming, deep forest green aesthetic with a pulsing orb that reacts to the AI's voice.

You simply set up your tea table, turn on your camera and microphone, and speak naturally. Master Chady watches your physical setup through the live video feed, guides you step-by-step through the Gongfu brewing process, sets mental intentions, and dynamically calls specific tools to provide accurate steep times and water temperatures based on the exact tea you are drinking.

How we built it

  • Frontend: Next.js 15 App Router, React 19, and Tailwind CSS.
  • AI Architecture: We used the official @google/genai SDK to connect to the Gemini Live API (gemini-2.5-flash-native-audio-preview-12-2025).
  • Multimodal Streaming: The browser captures microphone audio and video frames and streams them directly to Gemini Live over WebSockets via an ephemeral token. We specifically avoided proxying media through our backend to guarantee the lowest possible latency for a natural conversation.
  • Function Calling: We implemented Gemini Tool Calling so Master Chady can trigger UI events (like starting a visual brew timer) and fetch structured data for different tea profiles.
  • Infrastructure: The application is fully containerized and deployed on Google Cloud Run using a source-based build pipeline, ensuring it scales automatically.

Challenges we ran into

Achieving a production-grade deployment with a custom domain was our biggest hurdle. Initially, we faced severe routing issues trying to proxy our Google Cloud Run instance through Cloudflare's strict SSL headers. We eventually pivoted our architecture to deploy natively to the us-east4 (Northern Virginia) region to utilize Google Cloud's built-in Domain Mappings, allowing us to serve the app securely via chagather.com with zero proxy interference.

Accomplishments that we're proud of

  • The "Invisible UI": Breaking away from the standard "chatbot text feed" to create a genuinely immersive, distraction-free environment using CSS backdrop filters and ambient animations.
  • Sub-second Latency: Successfully maintaining a fluid, interruptible, two-way audio stream with video context without crashing the browser.
  • Authenticity: Translating real-world, authentic tea house expertise into a distinct, grounded AI persona that doesn't hallucinate brewing instructions.

What we learned

We gained massive insight into real-time WebRTC and WebSocket architectures. Handling ephemeral tokens securely while maintaining a live, bi-directional audio stream taught us how to optimize React re-renders so the main thread never blocks the audio playback.

What's next for ChaGather: Mindful AI Tea Companion

We plan to introduce a fully rigged 3D WebGL model of Master Chady that utilizes MediaPipe hand-tracking to physically react to the user pouring water in real time. We also plan to integrate a commerce flow so users can seamlessly purchase the physical teas that the AI recommends during their sessions.

Built With

Share this project:

Updates