Overview
Mimi is a multimodal AI assistant concept created to help older adults at the Koreatown Senior and Community Center access services from home instead of needing to physically visit the center for every interaction. What began as a mobility challenge quickly revealed a broader service access problem.
Many seniors had to show up in person just to register for classes, check lunch availability, or ask basic questions. That created a high friction experience for people already dealing with mobility limitations, long transit times, limited English proficiency, and low confidence with digital tools. In many cases, seniors had to invest time, energy, and risk before even knowing whether a trip would be worthwhile.
We designed Mimi as a voice and chat based assistant that could guide seniors through services remotely. Rather than treating the solution like a generic chatbot, we framed it as a guided, supportive presence rooted in a behavior users already trusted: asking for help over the phone.
We produced a short video that follows James, a lonely senior living in Koreatown, as he navigates the center’s outdated system and discovers Mimi.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.
Mimi high-fidelity prototype homepage, helping seniors to intuitively find what they need at the Koreatown Senior Center.
Context
The Koreatown Senior and Community Center is a major community anchor, providing meals, classes, and social support to older adults in the neighborhood. But many of its systems are still heavily analog. Seniors often need to physically come to the center to register, ask questions, or check whether resources are still available.
That structure creates real strain. Many seniors live alone, face mobility and transportation barriers, or have limited English proficiency. Some arrive as early as 2 a.m. to hold a place in line. Others travel across the city only to be turned away because capacity is already full. What should feel like access to community instead becomes exhausting, uncertain, and discouraging.
A key turning point in the project came when we reframed the challenge around presence rather than attendance. The problem was not simply how to help seniors get to the center more efficiently. It was how to help them feel informed, supported, and able to act before making the trip at all.
The Koreatown Senior and Community Center.
1 in 4 Korean seniors in L.A. live under the poverty line, and many more lack access to local community.
Research
Our research combined desk research, stakeholder conversations, journey mapping, service analysis, and concept testing to understand how seniors currently access KSCCLA’s services and where the experience breaks down.
The biggest insight was that the problem was not just transportation. It was uncertainty. Seniors often had to commit time, physical effort, and emotional energy before knowing whether a class had space, whether lunch was still available, or whether they could even get the help they needed.
We also found that many seniors already relied on guided, human support when navigating services. They often called family members, volunteers, or staff for help. That became a foundational insight for the concept: users were not rejecting support, they were rejecting systems that expected them to manage everything alone.
Key findings
In person registration and check in created major friction and contributed to churn.
Many seniors had limited English proficiency and low confidence with traditional web interfaces.
Existing digital experiences were not designed for older adults with accessibility, memory, and guidance needs.
Users were more comfortable with support that felt conversational, direct, and human.
Lack of visibility into service status made every trip feel risky and inefficient.
This pushed the project away from a standard website redesign and toward a guided access model that could reduce uncertainty before a senior ever left home.
Seniors waiting in line at the Senior Center.
1:30 a.m.
Sunja wakes up
1:45 a.m.
Eats instant ramen alone
1:50 a.m.
Walk to the bus stop in dangerous conditions
2:45 a.m.
Sit and wait for bus, hope it comes on time
3:00 a.m.
Catch bus and fend off dangerous bus riders
3:45 a.m.
Wait and hold place in line
4:45 a.m.
Continue waiting in line
10:30 a.m.
Wait for bus with no shelter
Process
The project moved through problem framing, journey mapping, concept ideation, storyboarding, wireframing, prototyping, and user testing.
Early journey mapping made one thing clear: much of the burden came from not knowing what would happen once someone arrived. Seniors were navigating high-effort situations with very little information upfront, often waking up in the middle of the night, commuting long distances, and waiting in line without knowing whether the outcome would justify the trip.
We used storyboards and speculative flows to explore what digital presence could look like for this audience. One early concept focused on allowing seniors to reserve lunch boxes in advance so they would know whether a trip was worth making. After speaking with stakeholders, we learned this would not work operationally. The center used a first come, first served model to avoid food waste, and unclaimed meals would create new problems. That feedback forced an important pivot. Instead of designing around reservation, we shifted toward real time lunch availability updates and better trip planning support.
As the concept evolved, the work became more focused on onboarding, conversation design, class registration, and the relationship between voice and screen. Each iteration helped us reduce complexity and move closer to a flow that felt familiar, manageable, and trustworthy.
Lunch storyboard
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.
Class Registration Storyboard
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 ranks her favorite classes to be chosen in the raffle.
Strategy
The core strategy behind Mimi was to design assistance, not just interface. We wanted the system to feel less like a tool users had to learn and more like a guide that could meet them where they already were.
Design principles
reduce uncertainty before travel
guide users step by step
support both voice and screen based interaction
build trust through clarity, familiarity, and visible support
keep help accessible at every point in the experience
This strategy directly shaped the multimodal approach. Audio alone was not enough. For tasks like class registration, reading out many options created too much cognitive load. But screen only flows also failed because they assumed comfort with typing, menus, and conventional navigation patterns. Mimi needed to combine voice, visual context, and progressive guidance so each mode could support the other.
We also intentionally grounded the experience in support patterns users already trusted, especially phone based help. The goal was not to make the service feel futuristic. It was to make digital access feel approachable, culturally aware, and emotionally safe.
The current Koreatown Senior Center homepage.
Conversation example between Mimi and an elderly man.
Validation
We tested the concept across multiple stages to understand how users reacted to Mimi and where the experience still created friction.
Early testing surfaced trust issues quickly. Users felt uncomfortable when asked for personal details too early, especially when it was unclear why the information was needed or how it would be used. Without context or trust cues, the experience felt intrusive.
At the same time, participants responded positively to being able to speak naturally or use talk to text, especially when typing felt difficult. They still wanted visual confirmation of what Mimi had heard so the interaction felt dependable.
Later testing revealed additional issues. Too many starting options created confusion, important next steps were easy to miss, and the site hierarchy felt cluttered. Users wanted larger type, fewer decisions, clearer class information, and easier access to a real phone number. A recurring theme was that even when users were open to digital help, they still wanted visible human backup.
Key refinements
simplified the homepage and reduced competing entry points
made key actions more visible and easier to find
reduced the amount of required information early in the flow
strengthened visual hierarchy, contrast, and readability
kept support pathways obvious, trustworthy, and close at hand
The final concept became much lighter, clearer, and better aligned with how seniors actually approach digital tasks.
Website homepage. Featuring the talk to Mimi service.
Talk to Mimi landing page. Options for calling on the web, text messaging, or calling on the users phone.
Calling Mimi on the web UI, featuring a chat box for conversation history.
Lunchbox registration page.
Website homepage concept featuring a simpler layout.
Second website homepage concept featuring more information.
Class registration page concept featuring a simpler layout.
Second class registration page concept featuring more visuals.
Solution
We designed Mimi as a multimodal AI assistant embedded into the KSCCLA website and broader service experience. Through voice and chat, it helps seniors check lunch availability, browse classes, sign up for programs, and get support without relying entirely on in person visits.
The final concept focused on making service access feel guided rather than transactional. Mimi responds to what the user is trying to do, understands what is on screen, and helps move them through tasks one step at a time. Instead of dropping users into a maze of pages or forms, the system creates a clearer path through common tasks and lowers the mental effort required to complete them.
Just as importantly, Mimi was designed as a bridge, not a replacement. It supports the center’s existing service model while giving seniors a more accessible and lower stress way to understand options, make decisions, and prepare before traveling.
High-fidelity prototype homepage.
Key Features
Homepage entry point
A clear, focused prompt introduces Mimi and what she can help with, making the starting point feel obvious.
Conversational guidance
Mimi supports both voice and chat, allowing users to choose the mode that feels most natural while receiving step by step help.
Class discovery and registration support
The assistant helps users understand available classes and move through registration in a guided way rather than navigating the site alone.
Lunch availability updates
Users can check current lunch status before deciding whether the trip is worth making.
Language first and accessibility focused onboarding
Mimi supports different comfort levels with language and technology through voice, subtitles, visual feedback, and simplified flows.
Persistent help and human backup
Support stays visible throughout the experience, including clear ways to call the center or get additional help when needed.
Together, these features positioned Mimi as more than a chatbot. It became a reassuring service layer built around real operational constraints, real user behavior, and real trust barriers.
Website homepage, featuring a pop-up screen to interact with Mimi.
Using the pop-up, the user can choose how they want to interact with Mimi. In this example, the user gets instructions on how to call Mimi on their phone.
Once the user calls Mimi, they input a code to sync the website UI with the call.
When a user wants to book a class, Mimi pulls up the class registration page and asks the user to pick the class they want.
Then Mimi pulls up the website calendar and asks the user which class date they would like to book.
Once the class is booked, Mimi pulls up a confirmation page and information on their booking. Once the users needs are met, they can just hang up the call and Mimi will close.
Results
The final Mimi concept tested significantly better than the existing website experience.
Key outcomes
100% of participants preferred the final Mimi flow over the current website
80% said booking classes with Mimi felt intuitive and was the simplest option
60% naturally chose Mimi instead of manually navigating the site
critical tasks like class registration and support became easier to self serve
the final flow felt more approachable and trustworthy for older users
More importantly, the project showed that conversational, accessibility focused AI could support service access in a way that felt practical, grounded, and genuinely useful.
Mimi access popup on the prototype homepage.
Impact
Mimi changed the frame of the problem from “how do we digitize registration?” to “how do we make service access feel available before someone arrives?”
That shift led to a stronger design direction. Rather than layering new technology onto an already confusing system, Mimi proposed a more supportive access model that helped users understand, decide, and act with more confidence. The concept suggests that thoughtfully designed AI can reduce friction for aging populations when it is grounded in real service conditions, real language needs, and real trust barriers.
At a broader level, the project points toward how community organizations can use multimodal assistance to expand access without losing the warmth and personal support that make those spaces valuable.
What I Learned
This project taught me that strong UX in service contexts depends on much more than interface clarity. It requires understanding emotional state, trust, language, mobility, cognitive load, and the broader systems around the experience.
One of the biggest lessons was how much stronger the solution became once the problem was reframed correctly. When we shifted from improving attendance to designing for presence, the project became sharper and more meaningful. That reframing opened the door to a solution that was not just more efficient, but more humane.
I also learned how important it is to test the assumptions behind a concept, not just the screens. Ideas that sound compelling in theory can break down once they meet real operational constraints, user hesitation, or trust issues. Storyboarding, prototyping, and testing were what made the project more grounded at every stage.
Reflection
Mimi is one of the projects that most clearly shaped how I think about design. It pushed me beyond interface work and into service design, accessibility, systems thinking, and emotional experience.
What made the project meaningful was not just the final prototype, but the way it responded to a real human problem with care. It reinforced for me that good design is often about reducing fear, confusion, and uncertainty just as much as it is about improving flows.
If developed further, Mimi could become a meaningful service layer for organizations that support older adults and other vulnerable communities, especially where digital access currently feels cold, confusing, or out of reach.
