Kindoc AI

Designing an AI-Powered Medical Checkup Chatbot

Overview

Kindoc AI helps users navigate the medical checkup journey. I led end-to-end UX and interaction design — defining the chatbot's role, IA, and key interactions, and working closely with our AI engineer to shape what the chatbot should be.

Role & Scope
UX research
UXUI Design
Timeline
2023.12~
Platform
Web
Mobile
Tools
Figma

Background

Users struggled throughout the entire checkup journey

We've consistently received feedback that users are not familiar with the terminology used in medical checkups. This has also increased the workload for our CX team.

SATISFACTION CURVE
↘ Satisfaction keeps sliding the further they go
DiscoverBrowseExploreBookCheckupResult
1 · Discover

Lands on the service's main page. Nothing in the way yet.

2 · Browse

There are just too many options, and it's hard to judge checkups you've never heard of — so people feel a bit lost.

3 · Explore

People don't fully trust the AI's answers yet, so they end up calling the clinic or CX team to hear it from an actual person.

4 · Book

Even after booking, people often circle back to the CX team to swap or change their checkup items.

5 · Get Checkup

Results arrive by mail — “Nice, my results are in!”

6 · Receive Result

People call in to understand their results, and clinics flag the same thing — there's a real need for a clearer way to explain what the results mean.

Tap or hover a point for what happens there

ROOT CAUSE

People can't really tell which checkups they need on their own — mostly because the terminology is unfamiliar and it's not clear what each one actually involves. So they keep reaching out to the clinic and CX team. It looked like something users should just figure out themselves, but it turned out to be a real load on our CX team too.

OPPORTUNITY AREAS

Build something that helps people make sense of the medical terms on their own — so they don't have to call the clinic or CX team just to understand what they're looking at.

Problem Keyword Finding

Low medical literacy hindered AI engagement

User feedback showed unfamiliarity with medical terms. I hypothesized this literacy gap would block AI engagement, and defined 'Limited medical literacy' as the core problem keyword.

TWO MOMENTS → ONE ROOT → TWO COSTS

② Browse Before the checkup

“Whoa, there are so many checkups — I’ve never even heard of most of them.”

⑥ Receive Result After the checkup

“I heard a quick explanation from the nurse, but I still wasn't sure — so I called to double-check what I'd heard.”

SHARED ROOT

Low medical literacy

COST TO USERS

They can't act on their own — not even sure what to ask.

COST TO US

Every unanswered question lands on the CX team.

Because users don't even know what to ask, we later seeded the chat with sample questions — turning a blank prompt into an obvious first step.

How Might We

Improve the overall checkup experience using an AI chatbot?

After summarizing the project background, I raised a How Might We question to clarify the project goal.

Design a conversational AI that supports users in understanding their health even when they don't know what to ask?

Desk Research

Researching AI capabilities and chatbot patterns

I conducted stakeholder interviews to understand our technical scope, then did competitive research on chatbot products to study proven UX patterns.

Stakeholder & Expert Interviews

To know what our AI Capability can provide

The interview helped me map the full chatbot journey and identify hallucination prevention as critical for credibility, shaping key features for users with limited medical literacy.

Competitive & Pattern Research

Defining Interaction Patterns to Adopt and Avoid

I analyzed how competitors structure onboarding and guide users through conversations. This shaped my approach to reducing blank-state anxiety and building a confidence-building flow.

Stakeholder & Expert Interviews
From interview with CTO, I found out preventing hallucination is vital and it's a crucial criteria in valuing a chatbot product.
Competitive & Pattern Research
Researched perplexity, gemini, clova and gpt.

UX Design Principles

Set up Principles and then iterated design.

I established core UX principles and made UI decisions throughout the entire journey in alignment with them.

  1. Guided example prompts

    Give people a way in when they don't know what to ask — the blank prompt becomes an obvious first step.

    Applied in — Final Delivery · Easy Start
  2. Follow-up tags

    Using backend data, the system offers up to five related questions so the thread keeps going after the first answer.

GROUNDED IN Both were scoped against data collected from five large companies — function-call coverage and result-sheet upload success rate.

Information Structure

Structured IA that meets both user and business sides.

The product started as a simple chatbot, but expanded to include educational content and clinic bookings. I carefully structured the IA to balance the core mission with business goals.

Business requirement Key feature
Global Navigation
Logo
Contents
Chat Section
SNB
New Conversation
Conversation History
Medical Service
Consult a Medical Professional
Settings
Language
Chat Workspace(Main)
Conversation Thread
Sample Questions
Question Input
File Upload
Upload Medical Report
Response Toolbar
Copy
Bookmark
Regenerate Answer
View Reasoning
External Bound
Reservation Website

*external landing

White box indicated initial purpose and pink ones indicate business needs.

Final Delivery

Equally provide great experience to every user.

Guided example prompts turn a blank input into an obvious first step.

Impact

Impact

These changes lowered the barrier to starting a conversation, made checkup results easier to understand, and left behind a reusable interaction framework for future healthcare AI services.

User Engagement

Reduced friction at the conversation-starting step. Improved usability and engagement of the result-interpretation features. Established a reusable interaction framework adopted by later healthcare AI services.

Business Expansion

Started as an internal R&D project and officially launched as Kindoc AI. Expanded into a Pro version, forming a product lineup. Adopted into the B2B2C checkup platform and a hospital-facing follow-up monitoring solution (PMS).