What is Nexus? A personal AI assistant that lives in your messaging apps, no screens to design, no buttons to place. Just conversation. The challenge: how do you design an experience when the entire interface is text? This project explores the emerging design space of human-AI interaction through a hands-on, iterative approach — designing, building, and living with a conversational AI companion.
Designing a traditional digital product starts with screens — wireframes, layouts, navigation patterns. But what happens when there is no screen? No buttons. No visual hierarchy. Just a blinking cursor in a messaging app. Nexus presented a fundamentally different design problem. The entire product experience exists within a conversation. Every interaction, from a morning news briefing to a complex career discussion, happens through text in WhatsApp, a platform I don't control and can't customize.
This constraint forced me to confront a set of questions that sit at the frontier of human-computer interaction:
How should an AI communicate across contexts?
A personal message is different from a group chat. A proactive alert is different from a response. The same AI needs to modulate its behavior based on who's listening, what's happening, and whether it was asked to speak at all.
What should an AI remember, and what should it forget?
Every conversation generates data. Names, preferences, decisions, offhand comments. Deciding what gets stored, how it's organized, and when it expires is a new form of information architecture — one where the consequences of getting it wrong aren't a confusing nav menu, but a violation of trust.
How much autonomy is too much?
An AI that asks permission for everything is useless. An AI that acts without asking is dangerous. The design challenge is finding the line — and making that line feel natural, not bureaucratic.
How do you build trust with something that isn't human?
An AI that asks permission for everything is useless. An AI that acts without asking is dangerous. The design challenge is finding the line — and making that line feel natural, not bureaucratic.
The invisible interface is the most personal one. Designing a conversational AI stripped away every tool I'd relied on as a designer and replaced them with something I'd never been trained to design: a relationship. Without screens, the AI's ability to read the room is the user experience — who's in the conversation, what was said before, whether the moment calls for depth or brevity. The quality of the experience isn't what the AI says. It's when, where, and whether it says anything at all.This revealed that memory and trust are design materials, not technical features. What an AI remembers shapes how the user feels — too little feels hollow, too much feels like surveillance. And capability without transparency erodes trust fast. The moment the AI surfaced something unexpected, my first reaction wasn't gratitude — it was "how did you know that?" That became a design principle: users need to see the seams.When the product has no visual form, the relationship becomes the product. And the tolerances for getting a relationship wrong are tighter than getting a screen wrong.