Mimi

A multimodal AI agent built to enhance service access for Koreatown seniors.
Project
Mobility Challange, SMC IxD
length
12 weeks
year
2024

Overview

Brief

Create an interactive prototype, to solve a mobility issue specifically affecting one neighborhood in Los Angeles.

Process

Stakeholder Interviews, Field Research, Journey Mapping, Storyboarding, Concept Ideation, Wireframes, User Testing, High-Fidelity Prototyping, Video Narrative Prototyping

Role

UX/UI Design, Research, Prototyping & Testing, Acting

Team

Austin Gregory, Oliver Jack, Yooji Chae, Princess Rivia

Problem

The arduous process of registration and check-in at KSCCLA causes many seniors to churn. Seniors are required to physically come in for every interaction, often waiting for hours in chaotic conditions. This prevents access to critical services like free lunches and social programs and increases the isolation of already vulnerable elders.

Insight

Physical presence is currently the only way to access information and take action at KSCCLA. But we realized that many seniors are already comfortable asking for help via phone. We hypothesized that a virtual assistant mimicking the warmth of a support call could reduce churn, especially if it worked across voice and screen interfaces. By “teleporting” into the center digitally, seniors could get the information they need, make decisions, and reduce trip anxiety.

Before: Seniors come to the center in person

After: They can now register from home

Solution

We created Mimi, a multimodal AI assistant designed to help seniors sign up for classes, check lunch availability, and navigate the system from anywhere. Mimi offers voice and chat based support, acting as both a guide and a companion.

Deliverables

High-Fidelity Prototype

We created a fully interactive prototype that simulates the real KSCCLA site with Mimi integrated. Seniors can chat, call, or navigate with Mimi to register for classes, check lunch availability, and ask for assistance. The prototype highlights voice interaction, visual feedback, and adaptive flows tailored for accessibility and language support.

Explore Prototype

Expressive Video Narrative

We produced a short video that follows James, a lonely Koreatown senior, as he navigates the center’s outdated system and discovers Mimi. The video demonstrates how Mimi helps him explore classes, receive real time lunch updates, and rekindle his joy for dancing, all without needing to stand in line or struggle with forms.

Background & Challenge

The Koreatown Senior and Community Center (KSCCLA) serves as a lifeline for many older adults in Koreatown, providing free meals, classes, and community. But its systems are overburdened and entirely analog. Seniors are required to physically show up for every task, whether registering for classes, checking lunch availability, or asking questions.

Many seniors live alone, exprience limited English proficiency, and have mobility constraints. Some arrive as early as 2 a.m. just to hold a spot in line. Others arrive too late, after hours of travel, and are turned away. This rigid system increases isolation and discourages participation.


The Koreatown Senior and Community Center.


Seniors waiting in line at the Senior Center.

Research & Discovery

Desk Research:
  • Only 4.82% of Koreatown residents have access to good parks, KSCCLA fills that role.
  • 1 in 4 Korean seniors in L.A. live under the poverty line.
  • 81% have limited English proficiency.
  • Waitlists for subsidized housing span 10+ years.

1 in 4 Korean seniors in L.A. live under the poverty line, and many more lack access to local community.

Journey Research

We had a hypothesis that the need for physical mobility at the center was causing churn. The class registration journey looks like this.

Journey Map

Journey mapping showed many seniors begin their trip by 1:30 a.m. KSCCLA volunteers manage up to 100 people on class registration days. Seniors are turned away due to overcapacity and limited visibility into system status.

1:50 a.m.
Walk to the bus stop in dangerous conditions

Article: LAPD seeks scooter rider in fatal hit-and-run of elderly Korean-American man

2:45 a.m.
Sit and wait for bus, hope it comes on time

Article: Buses regularly come 30, 40 minutes late

3:00 a.m.
Catch bus and fend off dangerous bus riders

Article: Seniors feel terrified on Metro, facing dangers with no staff in sight

3:45 a.m.
Wait and hold place in line

Article: The war of reception at KSCCLA

3:45 a.m.
Wait and hold place in line

Article: The war of reception at KSCCLA

10:30 a.m.
Wait for bus with no shelter

Article: In L.A's Koreatown, a lack of bus shelters debilitates elders

Insights:
  • The stress of in person processes is pushing seniors to stop attending entirely.
  • Seniors regularly call children or volunteers for help navigating technology.
  • A majority rely on analog systems because the current digital alternatives are not accessible.

Reenactment of a potential seniors trip to the senior center. His plan for the morning and walking in unsafe conditions to the bus stop.

Framing the Opportunity

We hypothesized that presence, not attendance, was the real need. What if seniors could "teleport" to the center digitally to make decisions, sign up, or learn what’s going on?

The center knows there’s a real need for a way to be present digitally. But they also shared that many of their seniors deal with dementia or just aren’t comfortable with technology. The current website is built more for typical web users, not for older adults with unique needs. Right now, seniors often rely on calling in, visiting in person, or asking family for help, none of which is really supported by the current site experience.

"Maybe in the future we have online registration on the homepage... But, but that is very, uh, difficult to do... Most of our seniors don't want to use the computer. And they can't use the computer."

- KSCCLA staff on the biggest thing they'd change about the sign-up process.

How Might We

How might we rethink presence at the KSCCLA, making information and action available from afar, so that seniors travel more intentionally?


Current Korea Town Senior Center homepage

Design Strategy: Introducing Mimi

We imagined Mimi as a friendly, AI-powered assistant that could recreate the comfort of a support call, something seniors already trust. We knew that digital presence didn’t just mean being online, it meant being guided. So we focused on making Mimi accessible through multiple modes: phone, chat, and voice.

Lunch storyboard

We sketched out an early idea where seniors could reserve a lunch box ahead of time, so they’d know the trip to the center would be worth it. The goal was to reduce stress and uncertainty by giving them a guaranteed pickup window.

But after speaking with the center, we learned that this wasn’t realistic. They prefer a first come, first served system to avoid waste, any unclaimed boxes would spoil. So we pivoted away from reservations and instead focused on giving users real time updates about availability.

In this experience, there's a time block every morning for reserving lunch.

Here, Sunja places her order the night before.

We explored a daily raffle system: winners could reserve a lunch slot, while others got updates and advice on whether it was worth making the trip.

Precedents

We drew inspiration from tools like SK Telecom’s ARIA, speculative UIs like Apple’s Knowledge Navigator, and the conversational warmth from the film Her. These helped inform a tone and structure that was both futuristic and grounded in emotional support.

We looked to SK Telecom’s ARIA, which lets seniors use voice commands to call friends and play games.

Apple’s old concept, the Knowledge Navigator, inspired us with its vision of a voice driven operating system.

the film Her helped shape our thinking around a friendly, goal oriented AI that feels personal and intuitive.

Conversation Design

We started thinking, what if Sunja didn’t have to go to the center every three months just to sign up for classes? What if she could do it from home?

That led us to a find first approach. If Sunja says she wants to book a class, Mimi needs to understand that and respond accordingly. That means the system has to support not just chat, but also account creation and easy access to the full list of available classes.

Multimodal Design

We found that audio only interactions, especially for tasks like filling out forms, can be really frustrating. People can only hold about 7 items in memory at once, so reading out all 42 class options just doesn’t work.

That’s why we focused on making Mimi context aware. She understands what the user is seeing on screen and can guide them based on where they are and what they might be trying to do next.


Conversation example between Mimi and an elderly man.

Registration Storyboard

Designed to analyze how seniors would move through the system.

Clerks advertise to seniors about the digitization of the queueing process.

Sunja onboards and chooses her language through audio.

Sunja onboards through natural language.

Sunja ranks her favorite classes to be chosen in the raffle.

Sunja receives a notification that she got into one of her favorite classes after the fact.

Onboarding

We designed Mimi to lead with audio, listing out options that users can either respond to by speaking or selecting on screen. This makes the experience more approachable for seniors who might not be comfortable with typing.


Audio website onboarding.


We also explored how Mimi could adapt for users with memory challenges, like dementia. The interface would shift based on the flow of conversation, and Mimi could offer gentle prompts or alternative paths if, for example, a user didn’t remember their address.


This idea was a lot like Her, an assistant that understands your routine and adjusts accordingly. Mimi could do something similar, using context to filter choices and shape the interface based on each person’s needs.

Iterative Design Process

Early Prototypes

Our first prototype explored how users would discover and interact with Mimi starting from the homepage. The landing page introduced Mimi as a friendly assistant ready to help with tasks like lunch updates, class registration, and general support. A clear message invited users to “Try out our new virtual assistant, Mimi,” reducing ambiguity and encouraging exploration.

Clicking this prompt led to a dedicated page that explained what Mimi could do. We outlined her core capabilities, like checking lunch availability, guiding class sign ups, and answering questions in Korean or English, to build confidence and familiarity.

From there, users were given two simple options: Chat with Mimi or Call Mimi. Whether the user preferred typing, speaking, or hearing a voice, Mimi was designed to meet them where they were most comfortable.

Choosing either option would lead to the conversational UI, where Mimi would guide the user through their request step-by-step. This early flow helped us test how seniors reacted to a multi-step introduction versus being dropped straight into a chatbot, and gave us insight into what kind of framing built the most trust.

Lunchbox registration page.

Early Prototypes User Testing

During testing, we found that users were uncomfortable being asked for personal details, like their address, apartment number, income, or documents right at the start. Without clear explanations or visible signs of trust, it felt intrusive. Testers wanted to understand why certain questions were being asked and what the information would be used for.

On the upside, they really appreciated being able to respond by voice or using talk-to-text, which felt more natural, especially for those who weren’t confident with typing or navigating menus. That said, people still wanted visual confirmation that what they said had been heard correctly.

The homepage also felt overwhelming, too many options and no clear focus. This made it hard for users to know where to start. Simplifying the layout and strengthening the visual hierarchy became a key priority moving forward.

Mid-Fidelity Prototypes

In this phase, we focused on refining the class sign-up flow based on the feedback from earlier testing. Users wanted fewer steps, larger text, and more clarity about what classes were available and how to register. We responded by streamlining the interface, improving the visual hierarchy, and emphasizing key actions.

The prototypes shown below illustrate Mimi guiding users through the class registration process, confirming their interests, helping them choose the right options, and making the experience feel personal and supportive.

Overall, this round of design aimed to make registration easier, faster, and more intuitive, while maintaining accessibility through both voice and visual inputs.

After users submitted their class selections, they are prompted to create a profile. We tested a Claude style interface where Mimi could automatically fill in the chat with helpful details and documents based on what the user had already shared.

User Testing

We ran another round of testing with three seniors, two of whom were Korean. One big takeaway: we need a smart default.

Right now, we offer three different starting options, one with audio, one without, and one that lets you explore freely. But the variety ended up confusing users. They weren’t sure where to begin and often didn’t engage at all. It made us ask: how can we make it feel more like something they already know? Users also had trouble finding the floating action button that’s supposed to help them finish class sign-up. Because of that, they missed the next steps entirely.

Another recurring issue was that basic info, like class types, dates, pricing, and contact details, wasn’t clear or easy to find. Everyone wanted to know what’s being offered, how much it costs, and how to join, without clicking through tons of pages. Across the board, the font was too small, the visual hierarchy was weak, and there was just too much going on. It made the site feel cluttered and tiring to use.

The sign-up process also felt too long. Most people just wanted to give their name, phone number, and maybe an email. They didn’t want to fill out a whole application just to show interest, and one thing that came up again and again: they wanted to talk to a real person. The lack of a visible phone number made people uneasy, especially those who don’t trust chatbots or online forms.

Quotes from user tests
“I would go to Google, call them up, and ask what they offer, what the price is, and what the time is, all that by calling them.”

- Korean User: W.

“And the other thing we talk about sight impairment or challenges. Don’t forget the hearing challenge as well. If you start doing audio stuff, you’ve got to realize there’s going to be a way if we can’t punch it up on our device, we’ve got to be able to hear you.”

- American User: F.

“You say the talk to someone, right? You need to have a microphone to speak to someone, but most of the senior, they don’t know how to make a microphone. That’s not easy, or the call on the phone. I don’t know how to do it…”

- Korean User: A.

Final Design

Feedback Improvments

For the final prototype, we focused on simplifying the experience and making key actions more obvious. We cleaned up the layout, removed distractions, and made the flow for meeting Mimi much more intuitive.

Right away, users are clearly shown what Mimi can help with, and her phone number is front and center, no more digging around for support. We wanted to reduce hesitation by making it feel like a friendly invitation, not a tech hurdle. Overall, the prototype feels lighter, more approachable, and tailored to how seniors actually navigate digital spaces.

Feature Highlights

Homepage Pop-Up

As soon as users land on the site, they’re greeted by Mimi through a clean, focused pop up. The options are simple and direct:

  • Call Mimi for help using a phone
  • Chat with Mimi to begin a guided flow
  • Watch a how-it-works video for a quick intro
  • We wanted the homepage to feel like a conversation starter, not a maze of links.

Class Discovery Calendar

Seniors can browse available classes in a simplified calendar interface. Mimi can filter results through voice or chat based on time, type or past activity. The system also remembers previous choices and preferences to make future sessions faster and more relevant.

Lunch Box Updates

Instead of guessing or waking up at 2 a.m., users can now check how many lunch boxes are left before they leave home. Mimi gives clear status updates like “85 lunches left, line moving quickly” so users can make informed decisions.

Conversational UI

We implemented a Claude style chat interface where Mimi fills in forms by asking short, focused questions. She only asks what’s essential and uses progressive disclosure when more info is needed, helping reduce cognitive overload.

Onboarding

We took a language first approach to onboarding. Users can say “I speak Korean” to begin in their preferred language. The experience is fully supported with voice, subtitles, and visual feedback. It’s built to adapt to each user’s comfort level with tech and language.

Smart Defaults & Visual Accessibility

To reduce decision fatigue, we guide users down a recommended path instead of giving too many options upfront. Fonts are large, colors are high contrast, and layout is intentionally linear. It’s easy to see where you are and what comes next.

Persistent Help & Phone Support

A help button stays accessible on every screen, offering quick links to chat with Mimi, call the center, or view frequently asked questions. This supports users who feel stuck and want human backup.

Emotional Design

Mimi’s tone is warm, calm, and encouraging. Rather than system speak, she uses phrases like, “Let’s take this one step at a time,” helping build trust and lower the intimidation factor.

Together, these features make Mimi not just a functional tool, but a familiar, reassuring presence designed to meet seniors where they are.

Explore Prototype

Reflection

Mimi began as an AI assistant, but became something deeper, a bridge between digital systems and the analog world that many elders still live in. I learned that real innovation happens when you bring emotional intelligence to systems design.

Designing Mimi required not just UX thinking but service design, cultural sensitivity and storytelling. From 2 a.m. journeys to effortless class registration, Mimi proves that presence doesn’t always mean being there.

I’m proud that we built something that could genuinely reduce friction in a vulnerable person’s day and hopeful for a future where tools like Mimi can help volunteer run organizations scale, without losing their heart.

Next Project