Rel.ai
The AI that learns
from what you don't say.
The Bet
A relationship AI that only learns when you remember to update it will always fail. People forget to talk to it when things are good, and can't bring themselves to when things fall apart. I bet on a different input model entirely. Passive, lightweight, continuous.
The Work
Two design decisions shaped this product along the way.
Redesigning the input layer
I replaced the single chat box with a compound input system: quick check-ins that feel like a game, reconciliation cards that turn real-life moments into AI context, emotional prompts timed to catch you in the right headspace. The AI learns from the moments between conversations, not just the ones you remember to report.
Making the feedback feel human
An AI that knows a lot about you can feel invasive or cold. I designed the feedback layer to feel more like a reflection than an analysis. Surface patterns gently, ask before concluding, never make the user feel diagnosed. The tone of the output was as designed as the interface itself.
The Impact
Currently in beta. The input architecture is the core differentiator.
Role
Founding Designer
Type
Multi-modal Input