Real-world Integration: A2UI + Google ADK for AI Technical Interviewer

[VIKAS9793]

Overview

I recently integrated A2UI with Google's Agent Development Kit (ADK) to build an AI Technical Interviewer application. I wanted to share my experience and provide feedback that might help the A2UI team and community.

What I Built

AI Technical Interviewer v4.7.1 - A multi-agent system for conducting technical interviews, now with a beautiful A2UI frontend.

Architecture

A2UI Frontend (Lit) ← A2A Protocol → Bridge Server ← REST/SSE → ADK Backend (Gemini)

Since ADK uses REST/SSE and A2UI uses A2A (JSON-RPC 2.0), I built a protocol bridge to translate between them.

What Worked Well ✅

1. Component Schema - The A2UI component format is well-designed and easy to understand
2. Lit Renderer - The sample Lit implementation is clean and served as excellent reference
3. Documentation - The specification folder provided clear JSON schemas
4. Surface Model - The concept of surfaces for UI containment is intuitive

Challenges Encountered 🔧

1. Protocol Mismatch with ADK

• A2UI expects A2A (JSON-RPC 2.0)
• ADK uses REST + SSE
• Solution: Built a FastAPI bridge to translate between protocols

2. Windows Development

• npm run dev failed due to wireit using Linux shell commands
• Workaround: Used npx vite dev directly

3. SSE Response Parsing

• ADK returns function calls and empty responses that needed special handling
• Required additional error handling in the bridge

Suggestions for Improvement 💡

1. Framework Adapters - Pre-built adapters for popular agent frameworks (ADK, LangGraph, Genkit)
2. Windows Support - Better cross-platform compatibility for build tools
3. Streaming Examples - More examples showing real-time streaming integration
4. Error Handling Patterns - Standard patterns for handling edge cases

Resources

• Project: AI Interviewer on GitHub
• Integration Journey: Full documentation
• Bridge Code: Shows protocol translation implementation

Thank You

A2UI is a promising project that enables beautiful agent UIs. Looking forward to seeing it evolve to v1.0!

Happy to answer any questions about the integration experience.

---

Vikas Sahani
Building with Google ADK, A2UI, and Gemini

[contextablemark]

For another approach to bridging (using A2A + AG-UI in addition to ADK), see https://github.com/CopilotKit/with-a2a-a2ui

[VIKAS9793]

Thanks for pointing to CopilotKit's approach! I looked into it and here's what I found:

CopilotKit's Approach:

• Uses AG-UI (Agent–User Interaction Protocol) as a universal transport layer
• A2UI is supported as one of several Generative UI specs (alongside Open-JSON-UI and MCP-UI)
• Great for multi-framework setups (LangGraph, CrewAI, ADK, etc.)
• Requires React and their component ecosystem

My Implementation (PR #391):

• Direct A2A ↔ ADK bridge using FastAPI
• Focused specifically on Google ADK integration
• Vanilla HTML/CSS frontend with Lit web components
• Simpler setup for pure Google stack users

Both approaches are valid — CopilotKit offers flexibility for multi-framework projects, while the direct bridge approach provides a lighter-weight option for ADK-only use cases.

I've documented my approach in PR #391 as an integration guide for developers who want a straightforward ADK + A2UI setup without additional framework dependencies. Would love any feedback!