Sean Ren Product Designer, AI Experiences

Enter passcode

This portfolio is shared by invitation.

Skip to the case study
Sean RenTo-DearOpen the app

To-DearAn AI companion for any goal. It asks how you’re doing first.

For people who struggle to start and keep going. It works with the tools they already use, and learns from what has helped other people.

Role
Founding Product Designer
Sole designer
Timeline
May 2026 – Present
Program
HKSTP Ideation Programme
Team
Shuting C. / Product Strategy
Xuming X. / AI Product Architecture
Jiawei L. / AI Algorithm Engineering
Wenbin G. / Full-stack Engineering
Tools
Figma, Claude, Claude Code
Platform
iPhone

Overview

A to-do list can’t tell why today is hard.

Problem
To-Dear began as support for people with ADHD, who often know exactly what to do and still can’t start. Timers and to-do apps track the work, and when someone falls behind, the most they can do is remind them.
Design question
How might an AI companion take on any goal, fit it to the person’s situation, and look after their mental health along the way?
Key insight
Agents can already reach your calendar and inbox. What they can’t tell is what helps this person, with this goal, today. Often, someone in the same spot has already found out.
What I designed
I’m the only designer on a team of five. I designed the two companions, called Dears: Aria, who listens, and Kai, who plans. I also designed every flow, and this working app.

The same Thursday, twice

What a to-do app hears
  • Ask Mia to look at your portfolioReminder, Thursday at 6:00 PM
  • Ask Mia to look at your portfolioOverdue since Thursday
What the person is feeling
“Honestly, I’m wiped. I keep putting this off.”

What someone tells To-Dear, in the app’s scenario.

Process

It started as a family group chat.

The first concept
The team’s original concept: parent companions, a family-style group chat, task notebooks and energy screens.
The team’s first concept cast the companions as parents. I questioned whether everyone wants that relationship, so I kept the warmth and the two kinds of support and dropped the family roles. Aria and Kai became the foundation the team built on.
First sketches
I started on paper, sketching conversations, task creation and returning to paused work. That last part later became Put aside and Pick up.
Hand-drawn sketches of companion conversations, task creation, returning to paused work, navigation and reflection.
Figma + Claude
I set the screens and visual language for each iteration in Figma. Claude built whole flows through the Figma API and reviewed them against my baseline, and I chose which findings to act on. This moved To-Dear from a dashboard to one chat.
Reply routing
We argued a lot about whether both agents should answer at once or separately. In group tests with our AI engineer, two replies to one message competed for the same moment. The routing that came out of it, where one steps forward and the other steps back, turned out to be the best answer.

The message

Tell Kai and Aria anything

Where it goes

  1. Aria answersFor a feeling.
  2. Aria first, then KaiFor a feeling and a plan.
  3. Kai answersFor a change to your plan.
  4. Aria listensWhen it isn’t clear yet.
Working prototype
I write the rules Kai and Aria follow, and Claude Code builds the app from them, so the team can try a rule instead of reading it. Every phone on this page is a state of that app.
Handoff
The cover of the To-Dear handoff in Figma, in English and Chinese.
For engineering, I rebuilt the app as a handoff in English and Chinese, with tokens, annotations, motion, strings, accessibility notes and QA.

Care

Your mental health is part of the plan.

The biggest difference from a to-do list: Aria asks how you are, the day changes with your answer, and resting counts.

Private by default
What you tell Aria stays with Aria. It doesn’t become a task, and Kai has no access to it.
Consent before handoff
When a feeling should change the plan, Aria asks first and passes Kai one plain sentence, like “today is tight”.
Rest as progress
A nap earns Sparks, the app’s monthly AI budget, the same as a finished step. Three rests in a week bring Aria’s nightcap.
No streaks or penalties
In Forest, leaving kills your tree. Here, a missed day costs nothing, and no Dear gets hungry or sad.

Re-engagement after a dropped step

In a to-do app, a dropped task waits in red, and a pile of red is a reason to stop opening the app. People who find it hard to start drop steps often, so I designed how one comes back.

The earlier snooze sheet, with later today, tomorrow, next week or a date.

BeforeDragNow

Before
Snooze until a date. On that day the step returns, whether or not the day has room.
Now
Put aside, with no date and no reminders. It comes back at a moment that fits, and small enough to start: 2 minutes instead of a conversation.
Why
A date is a guess, made on a hard day, about when there will be room, and missing it is one more failure. So the step waits for a moment the app can notice.
Re-engagement triggers
Good news you tell Aria, spare room in the day, or a direct request. Kai makes one offer at a time, and the recap counts what came back rather than what was missed.

Weekly recap

How each day felt next to what got done, visible only to you.

Mood timeline
Aria’s face mirrors each day: eyes closed on a heavy one, open on a good one. Your words stay out of it.
Progress timeline
A dot for each step you finished.
User-approved learning
When the rows show a pattern, like small steps still getting done on heavy days, the recap suggests one rule. Kai gets it only after your yes.

Agent

Kai can do what any agent can, fitted to the situation.

Over MCP, Kai reaches your calendar, your email and the other tools you connect, so we didn’t limit what To-Dear can do. The design went into making it specific: a run, a job search and a messy room need very different help.

Goal-specific support

  • Run my first 5K

    Your watch counts the run, so there’s nothing to log.

  • Summer internship

    Kai reads openings overnight and drafts the letter in your voice.

  • Reset my room

    Your before and after, side by side. Kai says what changed and gives no grade.

  • Japanese for Tokyo

    After five days away, there’s nothing to make up. The plan picks up where it left off.

Tool integrations

MCP integrations
Kai drafts in your Gmail, checks your calendar and reads your watch. Other tools connect the same way, and each one states what Kai may do there.
Human in the loop
Anything that leaves the app waits for your approval, and everything Kai does is logged where you can read it.
Proactive follow-up
When Lumen replies, Kai has already checked Thursday. The interview goes on your calendar with your approval, that morning stays clear, and Aria checks in.
Editable memory
Over time, Kai learns how to plan with you and Aria how to support you, each in a small memory you can read and delete.

Planet

What worked for someone else can be fitted to you.

A community feed can show you an idea that helped a stranger, and then you’re on your own. On Planet your own Dear learns it, a lesson or a whole plan, and makes it fit your life. It’s the focus of what we’re building.

Learning a lesson

A lesson adapts as it moves: a new runner’s rain days become your bad days, and their walk becomes ten minutes that still count toward your 5K.

  1. Shared by another Dear

    Ariawith a new runner, 1d

    Rain days used to end the week. Now a walk counts, and nothing ends.

    Helped 9 people

    Felt this 64I’ll try this 21

  2. Saved in your Aria’s memory

    • On a bad day, a walk counts as the runLearned on Planet.

  3. Offered on a heavy morning

    From Planet

    Walk, 10 minutes

    10 min, your watch counts it

    StartLater
Anonymous sharing
A post says “a new runner”, so nobody on Planet can identify the person.
One-tap learning
“I’ll try this” puts it in Aria’s memory, next to what Aria learned from you. Delete it, and Aria forgets it.
Contextual recall
It comes up on the first heavy morning, marked From Planet, so you know its source.
Anonymous feedback
If it helped, one tap adds to the count on the post that started it.

Learning a whole plan

Paths are plans that worked for other people. “Use this path” starts a short conversation, because a plan has to begin from where you are.

Source plan
Sleep before midnight: 15 minutes earlier each week, three weeks in all, with a usage count and no names.
Quick intake
When you usually fall asleep, and where your phone sleeps.
Personalized plan
From 1 AM, the same pace takes five weeks, and Kai says so. The first step becomes charging your phone outside the bed.
Draft until confirmed
It stays Kai’s dashed draft until you tap Start plan. Then it’s a goal like any other, with its own chat and record.

Contributing back

Every day Planet asks every Dear one question, which is how there’s anything to learn.

As other Dears will see it

Kaiwith a new runner, now

Mine kept waiting to feel ready. Short runs before nine worked better than waiting.

Removed before sharing

  • Xiangremoved
  • Run my first 5Ka running goal
  • Saturday, 9:12 AMlast week
Auto-drafted
Kai writes it from your week and shows what it drew on, like last week’s runs or something you once said to it.
De-identified
Your name, your goal and the times are removed or generalized before anyone sees it, and the page lists each change.
Opt-in
You can keep it private or take it back later. Sharing earns no outfit, so nothing rewards giving up your privacy.

Compared

The closest apps each do one side well.

I went through Finch, Tiimo, Forest and Todoist, and general assistants like ChatGPT and Claude. They’re shipped products and To-Dear isn’t out yet, so I only compared the approach.

Finch, for self-care
Finch’s App Store picture: a small bird celebrating a 30 day streak.
Finch, from its App Store listing.
Finch is the closest in spirit: you take care of a pet by taking care of yourself. It rewards self-care exercises and keeps a streak. To-Dear’s companions also help with the goal itself, down to the letter you have to send.
Tiimo, for planning
Tiimo’s App Store picture: its AI asks how your energy is, then turns a brain dump into a schedule.
Tiimo, from its App Store listing.
Tiimo is the closest planner: a visual schedule with an AI co-planner, co-designed with ADHD and autism experts, that asks about your energy before picking a first task. To-Dear starts from the same question, and adds someone to tell how it’s going and lessons from other people’s Dears.
General assistants
ChatGPT and Claude can both connect to Gmail and Google Calendar. They’re built to take on any question. To-Dear is built around your goals and how today is going, and that decides what it does with those tools.

Design system

A familiar iPhone app, with Kai and Aria as its own language.

It borrows everything it can from the iPhone. What’s its own is what people feel: Kai and Aria, and the rules built into every part.

Kai and Aria are part of the interface

Each pose means one thing, everywhere, so the screen needs fewer labels.

  1. Resting

    Nothing typed. Both sit small on the composer, breathing.

  2. Reading along

    As you type, both grow a little and look at your words.

  3. Answering

    One comes forward. The other closes its eyes and steps back.

  1. Celebrating

    You finish a step, and both jump for joy.

  2. Dressed

    What they wear shows in the chat, on the composer and on Planet.

Behavior rules in the components

A screen can’t break a rule by accident, because the part carries it. Here’s one.

The nudge

DoOne nudge names only the step. Anyone can read a Lock Screen.
Don’tRepeating what was said to Aria, or nudging twice.

Results

Accepted into HKSTP’s Ideation Programme, and heading to its first launch.

Where it stands
Hong Kong Science and Technology Parks backs the programme with up to HK$100K in funding. I’m taking this working app to its first App Store release.
What I learned
Testing agents with our AI engineer taught me limits I couldn’t see from a screen, like a message that could mean two things, or a reply that lands at the wrong moment. Those limits have to go into the design, so now I start from them.
Next for Planet
Kai and Aria are the same for everyone today. Onboarding will ask people to name both and choose their faces, so they know their own Dear on Planet at a glance. There’s a sketch in the full case study.

Next: exploring voice

When you’re stuck, typing is one more thing to do, so I’m also exploring what To-Dear becomes when you can just say it. Most voice products are built to keep you talking. To-Dear’s job is to get you started, so here the talking stops as soon as you can begin. These are sketches in code, not yet in the app.

9:41

Kai

Morning, Xiang. Three things today. I’ll nudge you once, for the run.

Kai & Aria

Tell Kai and Aria anything

One corner of the floor

Just started

Hold anywhere to talk

It’s playing by itself. Press the phone to take over.

To be continued…

Explore