Retail Store Example

Demo

Natural language queries become interactive visualizations. Using a retail store network domain, this demo shows the A2UI pattern of Intent-Based Rendering with Composite Recursion.

Try:

Enter a question above

The AI will create visualizations to answer it

A2UI Pattern

Intent-Based Rendering

The LLM outputs structured JSON intents (not code). The frontend validates with Zod and recursively renders using IntentRenderer dispatch.

Composite Recursion

Layouts like tabs and stacks contain child intents. IntentRenderer calls itself recursively to render nested structures.

Capability Negotiation

The client declares supported widgets. The LLM only generates intents for capabilities the frontend can render.

Confidence Scoring

Uses a 4-part rubric: Direct Answer, Data Fit, Specificity, and Clarity (each 0-25%). Other options: log-probs, user feedback, or validation checks.