H1

H2

H3

H4

Para Large

Para Regular

Caption

Button Text

AI Chatbot for Hospital Navigation

AI powered chatbot to help visitors navigate the University of Michigan Hospital

How I designed an AI chatbot that made our student startup profitable

Finding your way around a massive hospital is frustrating, it is a data backed problem for the hospital. My team at ImmiHealth changed that with a cheap, easy and simple solution - an AI chatbot with geolocated QR codes.

The key highlight is how real-world constraints and user feedback shaped the design, leading to impact and business.

Go straight to the solution

Process & Impact

Problem Statement, Research

Next Arrow

Deployed MVP, Initial Impact

Next Arrow

User Testing & Redesigning

Next Arrow

Final Impact

A dashboard-style visual with three main metric panels arranged horizontally, showcasing the positive impact of the deployed product. The first row, titled “User Satisfaction,” shows a value of 88% for the MVP and 92% for the redesigns, upward trend denoted with a small upward arrow. The second row, titled “Navigation related inquiries,” shows a value of 65% before the MVP, going down to 45.5% with the MVP and further going down to 36% for the redesigns, downward trend denoted with a small downward arrow. The third row, titled “Late arrivals for appointments,” shows a value of 34% before the MVP, going down to 29% with the MVP and further going down to 27.5% for the redesigns, downward trend denoted with a small downward arrow. The fourth row, titled “Average Interaction Time after asking for help,” shows a value of 7 minutes before the MVP, going down to 4.5 minutes with the MVP and further going down to 3.5% for the redesigns, downward trend denoted with a small downward arrow.A dashboard-style visual with three main metric panels arranged horizontally, showcasing the positive impact of the deployed product. The first row, titled “User Satisfaction,” shows a value of 88% for the MVP and 92% for the redesigns, upward trend denoted with a small upward arrow. The second row, titled “Navigation related inquiries,” shows a value of 65% before the MVP, going down to 45.5% with the MVP and further going down to 36% for the redesigns, downward trend denoted with a small downward arrow. The third row, titled “Late arrivals for appointments,” shows a value of 34% before the MVP, going down to 29% with the MVP and further going down to 27.5% for the redesigns, downward trend denoted with a small downward arrow. The fourth row, titled “Average Interaction Time after asking for help,” shows a value of 7 minutes before the MVP, going down to 4.5 minutes with the MVP and further going down to 3.5% for the redesigns, downward trend denoted with a small downward arrow.

My Team

ImmiHealth is a student-led tech startup focused on making healthcare more accessible for immigrants and non-English speakers, tackling challenges like insurance and hospital navigation.

We've been partnering with the University of Michigan Hospital for over a year, solving various problems.

Our team has grown from 4 to 6 students, and I am the Lead Designer.

As students, we worked within constraints on time, funding and expertise

Problem Statement

The Univeristy of Michigan Hospital is huge, navigation is a problem.

2.9 million Sq m, 1000+ beds, 29 clinical departments

Increased wait time (~12-15 mins on average)

Stress & frustration (reported by 65% patients)

Reduced patient satisfaction (score was ~25-30%)

These factors make navigation difficult for patients - especially first-time visitors and non-English speakers.

"It takes visitors about 14 minutes to reach their destination; 7 minutes even after asking for help. And that’s just after getting to the right floor. Navigation is a real challenge here."

Patient Experience Team, University of Michigan Hospital

Hospital's Existing Solution

The hospital currently relies on a volunteer program where undergraduate students assist patients with navigation.

Graphic representation, human figures denoting volunteers standing at various places on a hospital floor plan

These volunteers are stationed at specific locations during set times to assist visitors with navigation.

Our Constraints

Due to HIPAA, the Hospital would not:

Let us conduct user research

Give us any user data

Let us integrate with their IT infrastructure

Hence our solution had to be:

Quick, easy (to deploy and use)

Free and stand alone

Feasible for us

Our MVP Solution

The MVP solution was a QR-code-accessible AI chatbot trained on hospital maps to provide instant navigation assistance without requiring human volunteers.

Graphic representation, human figures denoting volunteers standing at various places on a hospital floor planGraphic representation, phones representing chatbots placed at various places on a hospital floor planGraphic representation, human figures denoting volunteers standing at various places on a hospital floor plan

This was feasible to implement and saved time, effort, and gave visitors autonomy while navigating the hospital.

AI also helped in saving development time.

AI Icon

Why AI?

Business need: quick deployment helped keep the Patient Eperience team interested in us

User need: Cheap, effective, easy to use

User need: Supports translation to other languages

UI Designs

Illustration of our healthcare-themed web app interface for Immihealth. It features a clean, modern layout with a simple ‘where do you want to go prompt on one screen and the other screens displaying how direction will be displayed on the app. The interface has soft blue and green tones.Illustration of our healthcare-themed web app interface for Immihealth. It features a clean, modern layout with a simple ‘where do you want to go prompt on one screen and the other screens displaying how direction will be displayed on the app. The interface has soft blue and green tones.

What did I do when the hospital did not allow user research?

Illustration of one screen, displaying how direction will be displayed on the app. The interface has soft blue and green tones.

I looked at existing UX research on AI chatbots and found some guiding design principles for our chatbot:

Hold context from conversation

Should have 0 response time

Should have option to talk to human/get out of the chatbot flow

Should give warning before repeating an answer

Saying "I don't understand" is less frustrating than giving the wrong answer

Politely tell user to only ask for direction if they ask any other information

MVP Impact

A dashboard-style visual with two main metric panels arranged horizontally, showcasing the positive impact of just the MVP of the deployed product. The first row, titled “User Satisfaction,” shows a value of 88% for the MVP. The second row, titled “Navigation related inquiries,” shows a value of 65% before the MVP, going down to 45.5% with the MVP, downward trend denoted with a small downward arrow. The third row, titled “Late arrivals for appointments,” shows a value of 34% before the MVP, going down to 29% with the MVP, downward trend denoted with a small downward arrow. The fourth row, titled “Average Interaction Time after asking for help,” shows a value of 7 minutes before the MVP, going down to 4.5 minutes with the MVP, downward trend denoted with a small downward arrow.A dashboard-style visual with two main metric panels arranged horizontally, showcasing the positive impact of just the MVP of the deployed product. The first row, titled “User Satisfaction,” shows a value of 88% for the MVP. The second row, titled “Navigation related inquiries,” shows a value of 65% before the MVP, going down to 45.5% with the MVP, downward trend denoted with a small downward arrow. The third row, titled “Late arrivals for appointments,” shows a value of 34% before the MVP, going down to 29% with the MVP, downward trend denoted with a small downward arrow. The fourth row, titled “Average Interaction Time after asking for help,” shows a value of 7 minutes before the MVP, going down to 4.5 minutes with the MVP, downward trend denoted with a small downward arrow.

User Research on the MVP

The initial success of our MVP demonstrated its value, leading the hospital to grant us permission for user research.

This allowed us to conduct Wizard of Oz testing and ethnographic studies to better understand user behavior and refine our solution.

Findings

Yay!

Eye Icon

When visitors walked into the hospital, they saw our posters with the QR codes and scanned them to use the chatbot!

Yay!

Binoculars Icon

When stuck after going around the hospital on their own, visitors looked for our posters for guidance from the chatbot!

Naah

Location Pin Icon

When they would scan a poster in the hospital, the chatbot could not understand where the visitors were by just asking them.

Down Arrow indicating that the problem we faced gave rise to the 'how might we' statement in the next section

How might we enable the hospital visitors to effectively describe their location so that the chatbot can understand and support them?

Our Redesigned Solution

Posters in different locations had different QR-codes.

Graphic representation, QR codes with area names representing different QR codes placed in different areas at various locations on a hospital floor plan

Based on which poster's QR code the user scans, the chatbot knows where the user is.

Final Impact

A dashboard-style visual with three main metric panels arranged horizontally, showcasing the positive impact of the deployed product. The first row, titled “User Satisfaction,” shows a value of 88% for the MVP and 92% for the redesigns, upward trend denoted with a small upward arrow. The second row, titled “Navigation related inquiries,” shows a value of 65% before the MVP, going down to 45.5% with the MVP and further going down to 36% for the redesigns, downward trend denoted with a small downward arrow. The third row, titled “Late arrivals for appointments,” shows a value of 34% before the MVP, going down to 29% with the MVP and further going down to 27.5% for the redesigns, downward trend denoted with a small downward arrow. The fourth row, titled “Average Interaction Time after asking for help,” shows a value of 7 minutes before the MVP, going down to 4.5 minutes with the MVP and further going down to 3.5% for the redesigns, downward trend denoted with a small downward arrow.A dashboard-style visual with three main metric panels arranged horizontally, showcasing the positive impact of the deployed product. The first row, titled “User Satisfaction,” shows a value of 88% for the MVP and 92% for the redesigns, upward trend denoted with a small upward arrow. The second row, titled “Navigation related inquiries,” shows a value of 65% before the MVP, going down to 45.5% with the MVP and further going down to 36% for the redesigns, downward trend denoted with a small downward arrow. The third row, titled “Late arrivals for appointments,” shows a value of 34% before the MVP, going down to 29% with the MVP and further going down to 27.5% for the redesigns, downward trend denoted with a small downward arrow. The fourth row, titled “Average Interaction Time after asking for help,” shows a value of 7 minutes before the MVP, going down to 4.5 minutes with the MVP and further going down to 3.5% for the redesigns, downward trend denoted with a small downward arrow.

Fin

That’s all from my project for now.

As I am still working on this project, I keep uncovering more user insights and ways to elicit them. I keep making my designs better in a very hands-on way.

This is an ongoing project with the Patient Experience Team at the University of Michigan Hospital. I'll keep our progress posted here. Feel free to get in touch for a detailed review of this case study!

Money saving feature for your banking app

A money saving feature in a banking app to help undergrad students develop money saving habits

How I modeled a solution based on user feedback at each step

This is the design of a feature in your banking app that helps you build a money saving habit. While I started out trying to building a solution for undergrad students specifically, it ended up being a solution that could be used universally.

The key highlight is user feedback influencing design changes at each step.

See Prototype of Final Solution

Background research

I initially wanted to build a separate app that helps you build money saving habits.

I looked at

Other apps that already do this (direct competition)

Popular ways of saving money that people use

I realized

A lot of apps do this already, so the competition was dense

It’s intimidating to know that there’s so many features in these apps

They are one-size-fits-all; target users have spouses, kids, cars and pets

Takeaways

I had to make it for undergrad students, there was nothing specifically for them. The app needed to be as simple and straightforward as possible to fit in with their busy lives.

Should be less time consuming

Should require less mental effort

Preferably habit forming so it’s difficult to ignore

User Interviews

I interviewed 6 undergraduate students.
Link to research plan and questionnaire

My objectives were:

Validate my assumptions about undergrad students

Ask stuff I didn’t know about them

My research plan consists of my:

Research objectives

Questionnaire

My success metrics for the research

Logistical findings:

Why a feature in a banking app? Why not a separate app?

No one used budgeting apps - they either never used them or used but then stopped using them

Making another app may not be a solution, but everyone has their banking app, so a feature in that app may be the solution

What do you use more, banking app or budgeting app?

No one used budgeting apps - they either never used them or used but then stopped using them .Making another app may not be a solution, but everyone has their banking app, so a feature in that app may be the solution

Finding out about user behavior

I realized different users had different relationships with money. As a result, it looked like we needed 2 solutions for 2 different kinds of users.

Soft-control users

Some students preferred to have full control over how they spend their money and would like slight reminders to control their spending habits

Hard-control users

Others did not trust themselves with money a lot and would like something more strict when it comes to a money saving app

Ideation

I brainstormed several solutions which I thought could be useful for the end user. After running my ideas by some undergraduate students (I even tested some) I realized that some ideas were well received while some were not. Additionally, users who liked soft control liked ideas differently than the ones that preferred hard control.

All ideas sorted:

Paper Prototype Testing

Fixes and Iterations

Here’s a link to all the detailed feedback from testing paper prototypes and the final UI.

It talks about what this feedback taught me and the changes I made based on it. What follows is a shorter version of the same list.

Areas of Success:

Both methods did help people be more mindful with their spending habits.

Micro-interactions successfully worked as a tool for teaching.

Aesthetics were well appreciated

Areas of Improvement and the fixes made:

The text was a lot

Reduced the text, tested it to make sure it was okay

Users needed a skip button during the square meditation so there’s a way of getting out of that flow quicker, instead of waiting their second time onwards. They felt stuck.

I added a skip button to both flows

When getting people to approve the payment, I initially had a notification. People just swiped it to the side to make it go away.

This told me that people needed a learning curve to understand what will happen each time they make a leisurely purchase and that a notification wouldn’t suffice, it had to a be something that grabbed more attention. So I added a learning curve explaining the process and changed the notification to an alert.

There are 3 options, the third option called ‘surprise me’ selects for you any one of the other 2 options at random. People selected that without looking at the other 2 options. But we want them to look at the other 2 options to decide which one would suite them better. We want them to use ‘surprise me’ only if they can’t decide right off the bat after looking at the other 2 options.

Hence, The surprise me option only comes after you have tested at least one option.

One user said they did not want to see both options, they would prefer a survey that determined what kind of a user they were and decide an approach for them. This is part of my next steps.

The Solution

Some things to know beforehand:

Most apps can tell when you are spending on something essential like groceries or gas and something non-essential like movie tickets or Starbucks. This feature will ONLY kick in when a non essential expenditure is taking place.

Students see a button that says ‘Manage your spending habits’ on the home page of their banking app. Microcopy tells them it’s a new feature.

Once they click on this button, they are given 2 options to choose from.

See Prototype of Final Solution

For soft control users

Every time you are about to make a leisurely transaction, the app will make you do square breathing. This will bring you back to the present moment. Then you will be asked in a prompt whether you would like to really go ahead with this purchase. You have the option to say yes, no or dismiss the prompt.

For hard control users

Every time you are about to make a leisurely transaction (say $12) that is above a set limit (say $10), the app will ask you if you would like to spend a little extra (say $16) where the extra money will go directly to your savings account. You have the option to say yes, no or dismiss the prompt. Here, whether you say yes or no, you end up saving.

For the long run

There are no negative repercussions for choosing to not save money.

We cannot expect one to always be on their best behavior.

Spending a little on yourself is crucial for one’s well being.

Reward mechanism

After 3-7 times of someone choosing to save, the prompt says ‘You can spend this time around - you’ve been good so far, you deserve this one!’

This creates variable reward (Hooked mechanism) because you never know when the ‘you deserve it’ prompt is coming.

I hope that this helps users want to save more, merely for the satisfaction of randomly getting one of these ‘you deserve it’ prompts.

Next Steps

I am still working on this project.

Here are a few things I am currently working on.

Work on teaching the user that they will get a notification every time a payment is about to go through

Test effectiveness of the alert that we used to replace the notification, as users dismissed the notification in the paper prototype

Test if habit formation is happening like I hypothesized

Test if reward mechanism is working like I hypothesized

Need to do some research about the survey feedback and ideate for it’s implementation

See how to incorporate a habit forming mechanism for the meditation option so it doesn’t feel like the user is being suddenly asked to reevaluate payment decision right before the payment

User research

Initially, was just trying to find a solution to one of the most common problems of the modern world, ‘phone addiction’.

What I looked for while conducting interviews:

Do people view this as a problem? Or are people okay with the way things are?

What are people’s beliefs about phone addiction?

What have people tried to do about their addiction? What has worked? What has not?

I interviewed 20 people in the age range 18-35.

Click here for a tentative questionnaire from my research plan

What did I find?

This is definitely a problem that people don’t have an effective solution for.

The cycle:

Keeping a phone around is inevitable. Sometimes, it’s needs for WORK and productivity

Once they start using a phone, a lot of time is automatically wasted

Wants to reduce screen time but no method has worked yet

Some things people have said:

“Sometimes I pick my phone up for some sort of work and automatically start scrolling instagram before anything else. Before I realise, a half an hour has passed and I have forgotten what I picked my phone up for!”

17/20 people had this problem

“I often use my phone for work, it’s inevitable. But if my phone is in my hand, I’m on facebook!”

18/20 people had this problem

“I’m used to feeling guilty after wasting time really. So much so, I don’t even feel the guilt anymore.”

17/20 people had this problem

“I think about reducing my phone time every day, but nothing has worked so far and I doubt anything really will.”

20/20 people had this problem

The problem

It was easy to mindlessly hop onto apps like Instagram, Netflix and even Zomato and waste time without much thought.

Mindlessness was the problem, mindfulness could be a possible solution.

How might we make smartphone usage more mindful?

Ideation and brainstorming

These are thumbnails of the ideas I sketched out.

All these ideas were thought of from a perspective of ‘how can we make the user mindful that they’re using their phone’ rather than ‘ how can we stop the user from using their phone’.

After a thorough SWOT analysis, I narrowed it down to 1 solution.

Click to enlarge

Paper prototyping content to make sure it's not pushy, rather amicable.

The wording is carefully designed to feel amicable, not Intruding and pushing you to be mindful. It has also been thoroughly tested for the same.

Gina is an assistive character added to make it feel more personalized than digital.

You thereby save yourself from wasting time by rabbit holing on an addictive app!

Sitemap

Wireframes

UI

Fin

That’s all from my project for now.

As I am still working on this project, I keep uncovering more user insights and ways to elicit them. I keep making my designs better in a very hands-on way.

That’s all from my project for now.

As I am still working on this project, I keep uncovering more user insights and ways to elicit them. I keep making my designs better in a very hands-on way.

Checkout other projects

Backtrack Thumbnail

Backtrack

SampleServe Thumbnail

SampleServe UX Redesign

Let’s make designs that

stand out

show expertise

are humane

gain trust

Get In Touch