Liver Disease Prediction Using Machine Learning
Liver Disease Prediction Using Machine Learning
Liver Disease Prediction Using Machine Learning
UX Case Study
UX Case Study
UX Case Study
Role: UI/UX Designer (cross-functional collaboration in research, design, and development)
Organization: IBM – Academic Collaboration, India
Duration: Aug 2022 – Dec 2022
Tools: Figma, IBM Cloud, Python, IBM Cognos, HTML/CSS
Team: 4 members – cross-functional (UX, Development, Data Science)
Team ID: PNT2022TMID01272
Redesign: Independent redesign completed in 2025
Role: UI/UX Designer (cross-functional collaboration in research, design, and development)
Organization: IBM – Academic Collaboration, India
Duration: Aug 2022 – Dec 2022
Tools: Figma, IBM Cloud, Python, IBM Cognos, HTML/CSS
Team: 4 members – cross-functional (UX, Development, Data Science)
Team ID: PNT2022TMID01272
Redesign: Independent redesign completed in 2025
Role: UI/UX Designer (cross-functional collaboration in research, design, and development)
Organization: IBM – Academic Collaboration, India
Duration: Aug 2022 – Dec 2022
Tools: Figma, IBM Cloud, Python, IBM Cognos, HTML/CSS
Team: 4 members – cross-functional (UX, Development, Data Science)
Team ID: PNT2022TMID01272
Redesign: Independent redesign completed in 2025
Back in 2022, we developed a liver disease prediction tool using machine learning as part of an academic collaboration with IBM. While the model was accurate, the interface lacked empathy. In 2025, I revisited the project through a design-thinking lens transforming it into a clear, intuitive experience built for real clinical environments.
Back in 2022, we developed a liver disease prediction tool using machine learning as part of an academic collaboration with IBM. While the model was accurate, the interface lacked empathy. In 2025, I revisited the project through a design-thinking lens transforming it into a clear, intuitive experience built for real clinical environments.
Back in 2022, we developed a liver disease prediction tool using machine learning as part of an academic collaboration with IBM. While the model was accurate, the interface lacked empathy. In 2025, I revisited the project through a design-thinking lens transforming it into a clear, intuitive experience built for real clinical environments.
Defining the problem
Defining the problem
Defining the problem
We started with doctors. In many clinics, a liver diagnosis meant invasive biopsies painful, expensive, and often delayed. Meanwhile, lab technicians had their own hurdles: outdated tools with steep learning curves and poor accessibility.
What they needed was clear:
Fast predictions from basic blood test results
Zero-friction data entry
Confidence in the output
We started with doctors. In many clinics, a liver diagnosis meant invasive biopsies painful, expensive, and often delayed. Meanwhile, lab technicians had their own hurdles: outdated tools with steep learning curves and poor accessibility.
What they needed was clear:
Fast predictions from basic blood test results
Zero-friction data entry
Confidence in the output
We started with doctors. In many clinics, a liver diagnosis meant invasive biopsies painful, expensive, and often delayed. Meanwhile, lab technicians had their own hurdles: outdated tools with steep learning curves and poor accessibility.
What they needed was clear:
Fast predictions from basic blood test results
Zero-friction data entry
Confidence in the output
So, I reframed the design question:
So, I reframed the design question:
So, I reframed the design question:
“How might we make liver disease prediction feel less like a model and more like a tool doctors actually want to use?”
“How might we make liver disease prediction feel less like a model and more like a tool doctors actually want to use?”
“How might we make liver disease prediction feel less like a model and more like a tool doctors actually want to use?”
The first version
The first version
The first version
Our team, 4 members across UX, development, ML, and strategy built the first prototype in Flask and IBM Cloud. The model used routine LFT values like bilirubin and enzymes to output a simple yes/no prediction.
It worked well. But the experience was skeletal. You had to input raw numbers into a plain form, hit “predict,” and read a basic result.
There was no feedback, no flow and no reason to trust what the system said.
Our team, 4 members across UX, development, ML, and strategy built the first prototype in Flask and IBM Cloud. The model used routine LFT values like bilirubin and enzymes to output a simple yes/no prediction.
It worked well. But the experience was skeletal. You had to input raw numbers into a plain form, hit “predict,” and read a basic result.
There was no feedback, no flow and no reason to trust what the system said.
Our team, 4 members across UX, development, ML, and strategy built the first prototype in Flask and IBM Cloud. The model used routine LFT values like bilirubin and enzymes to output a simple yes/no prediction.
It worked well. But the experience was skeletal. You had to input raw numbers into a plain form, hit “predict,” and read a basic result.
There was no feedback, no flow and no reason to trust what the system said.
"How might I redesign a data-heavy prediction tool into a meaningful experience for clinicians and lab technicians one that’s fast, intuitive, and human?"
"How might I redesign a data-heavy prediction tool into a meaningful experience for clinicians and lab technicians one that’s fast, intuitive, and human?"
"How might I redesign a data-heavy prediction tool into a meaningful experience for clinicians and lab technicians one that’s fast, intuitive, and human?"
Rebuilding with empathy
Rebuilding with empathy
Rebuilding with empathy
In 2025, I took everything we’d learned and rebuilt the interface from the ground up.
This time, it wasn’t just about delivering output. It was about supporting decision-making — especially for doctors with limited time, and tech constraints.
I mapped out a new flow that included onboarding, predictions, results, and even follow-ups. I introduced smart features like chatbot integration, in-line explanations, and appointment booking.
And then I designed it to feel calm.
In 2025, I took everything we’d learned and rebuilt the interface from the ground up.
This time, it wasn’t just about delivering output. It was about supporting decision-making — especially for doctors with limited time, and tech constraints.
I mapped out a new flow that included onboarding, predictions, results, and even follow-ups. I introduced smart features like chatbot integration, in-line explanations, and appointment booking.
And then I designed it to feel calm.
In 2025, I took everything we’d learned and rebuilt the interface from the ground up.
This time, it wasn’t just about delivering output. It was about supporting decision-making — especially for doctors with limited time, and tech constraints.
I mapped out a new flow that included onboarding, predictions, results, and even follow-ups. I introduced smart features like chatbot integration, in-line explanations, and appointment booking.
And then I designed it to feel calm.
Wireframes: Setting the Foundation
Wireframes: Setting the Foundation
Wireframes: Setting the Foundation
Before refining the interface with a design system, I created an early wireframe to translate the tool’s functionality into a clear, navigable layout. While the wireframe was based on our initial structure (since updated in the redesign), it served as a critical blueprint for thinking through hierarchy, content placement, and user flow.
Before refining the interface with a design system, I created an early wireframe to translate the tool’s functionality into a clear, navigable layout. While the wireframe was based on our initial structure (since updated in the redesign), it served as a critical blueprint for thinking through hierarchy, content placement, and user flow.
Before refining the interface with a design system, I created an early wireframe to translate the tool’s functionality into a clear, navigable layout. While the wireframe was based on our initial structure (since updated in the redesign), it served as a critical blueprint for thinking through hierarchy, content placement, and user flow.
From Wireframe to Refined Flow
From Wireframe to Refined Flow
From Wireframe to Refined Flow
While the original wireframe laid a solid foundation, it also revealed friction points — extra steps, unclear navigation, and areas where users could get lost. Revisiting the structure gave me the opportunity to streamline the journey, reduce decision fatigue, and make critical actions faster to access.
While the original wireframe laid a solid foundation, it also revealed friction points — extra steps, unclear navigation, and areas where users could get lost. Revisiting the structure gave me the opportunity to streamline the journey, reduce decision fatigue, and make critical actions faster to access.
While the original wireframe laid a solid foundation, it also revealed friction points — extra steps, unclear navigation, and areas where users could get lost. Revisiting the structure gave me the opportunity to streamline the journey, reduce decision fatigue, and make critical actions faster to access.
TECHNICAL FLOW
TECHNICAL FLOW
Sitemap
Sitemap
Design system for clarity
Design system for clarity
Design system for clarity
Doctors don’t want clutter. So the interface needed to feel clinical, but warm.
Every element was chosen to support one thing: clarity under pressure.
Doctors don’t want clutter. So the interface needed to feel clinical, but warm.
Every element was chosen to support one thing: clarity under pressure.
Doctors don’t want clutter. So the interface needed to feel clinical, but warm.
Every element was chosen to support one thing: clarity under pressure.
A simple, no-fuss workflow
A simple, no-fuss workflow
A simple, no-fuss workflow
The interface was intentionally kept straightforward — guiding users clearly from one step to the next without overcomplicating things. No clutter, no confusion.
The interface was intentionally kept straightforward — guiding users clearly from one step to the next without overcomplicating things. No clutter, no confusion.
The interface was intentionally kept straightforward — guiding users clearly from one step to the next without overcomplicating things. No clutter, no confusion.
Appointment booking
Appointment booking
Appointment booking
I wanted scheduling to be as quick as noting something in your diary. The flow is simple: select a doctor, pick a date, choose a time, and confirm. No extra screens, no unnecessary steps just a fast, clear path to getting an appointment booked.
I wanted scheduling to be as quick as noting something in your diary. The flow is simple: select a doctor, pick a date, choose a time, and confirm. No extra screens, no unnecessary steps just a fast, clear path to getting an appointment booked.
I wanted scheduling to be as quick as noting something in your diary. The flow is simple: select a doctor, pick a date, choose a time, and confirm. No extra screens, no unnecessary steps just a fast, clear path to getting an appointment booked.
Chatbot & conversation
Chatbot & conversation
Chatbot & conversation
Medical tools often feel cold and transactional. I wanted this to feel more like a companion.
So I added a hybrid chatbot part rule-based, part NLP that could guide users through the result.
It wii answer questions like:
“What does ALT mean?”
“Is this a high value?”
“Should I refer this patient?”
And if a follow-up was needed, doctors could book an appointment right from the chat.
Medical tools often feel cold and transactional. I wanted this to feel more like a companion.
So I added a hybrid chatbot part rule-based, part NLP that could guide users through the result.
It wii answer questions like:
“What does ALT mean?”
“Is this a high value?”
“Should I refer this patient?”
And if a follow-up was needed, doctors could book an appointment right from the chat.
Medical tools often feel cold and transactional. I wanted this to feel more like a companion.
So I added a hybrid chatbot part rule-based, part NLP that could guide users through the result.
It wii answer questions like:
“What does ALT mean?”
“Is this a high value?”
“Should I refer this patient?”
And if a follow-up was needed, doctors could book an appointment right from the chat.
Designed for real-world access
Designed for real-world access
Designed for real-world access
This wasn’t built for cutting-edge hospitals. It had to function in semi-urban clinics with no EMRs, limited bandwidth, and basic devices.
So I made sure it worked seamlessly across manual data entry or lab integration.
Lightweight UI with fast load times no installs, just open and use
This wasn’t built for cutting-edge hospitals. It had to function in semi-urban clinics with no EMRs, limited bandwidth, and basic devices.
So I made sure it worked seamlessly across manual data entry or lab integration.
Lightweight UI with fast load times no installs, just open and use
This wasn’t built for cutting-edge hospitals. It had to function in semi-urban clinics with no EMRs, limited bandwidth, and basic devices.
So I made sure it worked seamlessly across manual data entry or lab integration.
Lightweight UI with fast load times no installs, just open and use
"It wasn’t just about inclusivity in tone — it was about infrastructure"
"It wasn’t just about inclusivity in tone — it was about infrastructure"
"It wasn’t just about inclusivity in tone — it was about infrastructure"
What I took away
What I took away
What I took away
Good design is about more than screens it's about advocacy.
Throughout this project, I found myself asking the same question again and again: Does this help someone do their job better? If the answer wasn’t yes, we changed it.
Because in healthcare, clarity isn’t nice to have it’s critical
Good design is about more than screens it's about advocacy.
Throughout this project, I found myself asking the same question again and again: Does this help someone do their job better? If the answer wasn’t yes, we changed it.
Because in healthcare, clarity isn’t nice to have it’s critical
Good design is about more than screens it's about advocacy.
Throughout this project, I found myself asking the same question again and again: Does this help someone do their job better? If the answer wasn’t yes, we changed it.
Because in healthcare, clarity isn’t nice to have it’s critical