The invisible interface: Designing GenAI conversations for three billion “offline” users

We often talk about “mobile-first” design, but what about “voice-only” design for users who have never used the internet? At Viamo, Victor led the design of a GenAI-powered service for 3 billion people in emerging markets, accessed entirely through basic mobile phones (IVR).

In this deep dive, he'll show you what happens when content design is the only UI. When users can’t see a screen, “navigation” becomes a conversation, and “error states” must be spoken with empathy, not just displayed.

He'll break down the specific prompt architecture we used to reduce AI hallucinations in health and finance advice, achieving a 30% increase in response accuracy. We’ll look at the “information architecture of sound,” how to structure conversational trees when users can’t “go back,” and how to write prompts that survive the journey from a sophisticated LLM to a low-fidelity phone speaker. This is a masterclass in extreme constraints and the ultimate test of clarity.

In this session, you’ll learn how to:

• Architect “invisible” navigation menus that users can hold in their working memory.
• Write system prompts that force LLMs to speak in “plain language” suitable for oral cultures.
• Structure error recovery flows for voice interactions where “refreshing the page” isn’t an option.
• Measure success using conversation depth and completion rates instead of clicks.