Product Design · AI Emotional Companion App
Bestie AI
Role
Product design: the visual side, the relationship feature end to end, and the subscription screens.
Duration
Dec 2025 - Jul 2026
Type
Mobile Product / AI-Native Features
Skills
Product Design, UI/UX, Monetization UX, Design Systems
Tools
Figma, Rive, AIGC tools
An AI companion's memory is only as meaningful as the surface that carries it. Bestie remembers the people you talk about; I designed the feature that turns that memory into something you can feel.
Bestie is an AI emotional-support companion with five personas, used mainly across Europe, the US and Japan. I also designed its subscription screens; this page leads with the relationship feature.


Shipped · the people you talk about, kept by the model and shown as an orbit
The model knows your people. What should the product do with that?
Bestie's model already extracts the people from your conversations, one profile per person, built automatically (a user can still add someone by hand, but doesn't have to). That capability has two failure modes: shown raw, it reads like a CRM; shown carelessly, it reads like surveillance. The design problem was to make machine memory feel like being known, not being tracked. Every decision below answers to that.
From a contact list to a felt universe
When I picked this up, the feature existed as a flat list of contacts, functional, and exactly the CRM failure mode. I designed what it became: you sit at the center, the people in your life orbit you, and each person opens into a profile the model keeps for you.
I came in early, with the goal defined but the shape of the feature still open: what a person is in this product, what the model keeps about them, how it all fits, and the form it takes on screen.
Before
After




One person at a time, or the people around you?

I sketched people as bubbles with detail cards underneath, then tried giving the people the whole screen. The cards made one person’s details easy to read, while the open layout left room for more people. In the final design, you find someone on the orbit, then open their profile to explore how you feel around them.
Three decisions that made it feel like Bestie
In the usage report, family was the most common profile category. Romantic profiles were more common among people with one or two profiles. People with larger collections also kept profiles for coworkers and acquaintances.
Who they are and how things are going
I kept identity and current state separate in the profile. Someone can still be your coworker after weeks without contact. The relationship label can stay the same while the description of your recent interactions changes.
When does a radar dimension unlock?
A profile the model writes on its own will sometimes be thin, and sometimes wrong. The design had to make both cases visible instead of papering over them.
The radar fills in as you talk
An axis appears only once there's enough conversation to read it. Until then it stays locked, and any number the model can't back yet shows as “--”. The more you talk about someone, the more of the shape you see.



An invitation to talk
From the start, the relationship universe was designed to let users decide what a relationship means to them. These three final pop-ups can appear while someone explores their relationships. Each offers a topic and an entry into chat with Bestie.



Relationship quizzes

The tools sit inside each person’s profile. In Sam’s Toolkit, the first card offers a test and the next lets you draw an energy card. You can pick whichever feels right in that moment.
Reading a mood while choosing it


Before the redesign · v1.2.7
In the early flow, people picked a mood in chat and revisited their entries by date in My Journey. For the redesign, I explored how someone could choose a more precise word with one thumb.
View the original check-in flow
One gesture moved from negative to positive. Two unpleasant feelings could land in the same place even if one felt quiet and the other intense.

Two directions captured pleasantness and intensity. Each position had three mood words, but a thumb covered the words just as the user needed to read them.

I moved the words above the palette and enlarged the touch area so people could read their selection while dragging.
Open the full design studiesWho gets the next turn?
In a chat with five characters, tapping an avatar to invite a reply sat close to the mute action. I worked on the panel feedback so people could tell whom they had called on. The selected character moves to the center, the speaking avatar pulses, and muted characters recede. I also designed how mute actions appear in the conversation: if no one has spoken between two mutes, the record groups both avatars in one line.



Subscription experience

Before I joined · Aura membership
Version 1.3.1 compared the free plan with Aura. People could choose annual or monthly billing and start a three-day free trial.
The usage-based redesign
Later, the team introduced four usage-based tiers to manage AI conversation costs. The PM set prices and limits. I designed the notices people saw in chat and on calls, and the path from each notice to the upgrade screen.
Same ask, shaped to the moment
The paywall reskins to whichever limit the user just reached: a soft banner in chat when the weekly quota runs out, and a full screen whose headline, trigger and unlocked features change with the feature — voice replies, live calls, or Bestie tasks.



Plans and offers
The membership screen compares Basic, Plus, Max and Unlimited. A separate offer screen is designed for returning subscribers.



What shipped
What I take from this
I keep coming back to the moment when someone reads a relationship prompt. They might have a story that changes how the prompt reads. I wanted them to be able to tell that story. The same care mattered when a call hit a limit: explain what ran out, then let the person decide whether to upgrade.
- Where it fell short
- I do not have a separate measure of how often a relationship prompt led to a conversation.
- Next iteration
- I would ask someone to read one radar axis in their own words and tell me whose experience they think it describes.
The most AI-native half of this role, writing the rules a generation pipeline must follow and designing its human-in-the-loop checkpoints, is a separate story: AI Creative Pipeline for Growth →