Conversation as a Signal Query Engine - HQIQ Maestro

Maestro / Hadi is a voice - first generative UI agent platform — no chat bubble, no text input. A full-screen avatar composes the interface in real time as the user speaks: scenes transition, listings render, comparison views compose themselves, contextual sheets open. The agent IS the frontend.

We have shipped Translator Mode — a conversational capability where the same agent enters bilingual interpreter persona mid-session when asked. A realtor speaks English, a guest speaks Arabic (or any of 30+ languages Gemini Live supports), the avatar mediates the exchange in real-time bidirectional audio, then exits gracefully when the realtor asks to return to normal — with all prior property context preserved.

Architecture: two function tools (enable_translator_mode / disable_translator_mode) inject interpreter instructions into the agent’s chat context via update_chat_ctx, and emit TRANSLATOR_MODE data packets that surface a UI status indicator on the frontend. The LLM detects the trigger phrase naturally — no regex, no state machine. Provider-agnostic: runs on Google DeepMind’s Gemini Live API for native audio, or OpenAI Realtime as a fallback.

Litmus test: “Would this have been impossible with a chat interface?” Yes. Voice translation IS the conversational cadence chat removes. The translator pattern generalizes — same hook supports accessibility narrators, interview coaches, any role-shaped persona switch. The agent’s persona is its capability, not its costume.