Cognitive State-Aware Adaptive Learning Interface
Redbio is a learning platform that reads a learner's state instead of asking for it.
Heart rate, brainwave activity and camera signals infer immersion and fatigue while a session is running.
I led the UX, turning those signals into changes the learner actually feels, in content, pacing and the interface itself.
ROLE
Lead UX Designer, Team of two
Client
Graduation exhibition,
EEG sensors and companion app
Timeline
July to December 2024
Contribution
Research, UT, IA, UX/UI direction
01 | overview
Key highlights
Every learning platform adapts what it teaches. None of them adapt to the person sitting in front of it.
Redbio reads immersion and fatigue while the session runs, then moves the content, the pace and the interface to match.
The result is a session that corrects itself, so a bad twenty minutes does not cost the whole hour.
Automatic Health & Mood Detection
Redbio uses camera and sensor data to detect the learner’s condition and automatically set the starting difficulty.
Adaptive Learning in Real Time
Redbio adapts content and interface in real time using stress levels, heart rate, and immersion.
Visualizing Your Cognitive Patterns
The post-session report visualizes immersion, fatigue, and stress to show when attention drops.
02 | research
background
Korean students have found a shortcut, and it is a controlled substance.
Methylphenidate is an ADHD stimulant that test-prep districts call the study-well pill.
It more than tripled in four years, and more than half of it now goes to people under twenty.
By 2026 it was no longer niche.
More middle and high schoolers had taken a prescription drug without a prescription than had smoked a cigarette, and methylphenidate was the most common one.
What they are buying is not a drug. It is focus that holds.
No learning platform sells that.
Where the pressure comes from
The pressure is not abstract. It has a shape.
Every year there are fewer students, and every year more money is spent on them.
It reaches one student as one thing: hours.
Problem and goal
Hours are only useful while focus lasts, and focus does not last as long as the hours do.
Every tool a student uses treats a tired learner exactly like a fresh one.
Four things go wrong because of that, and each one gave the project a goal.
How does a learner sustain focus and manage stress without reaching for a prescription?
💭 How does a learner sustain focus and manage stress without reaching for a prescription?
research methodology
Redbio only works if a learner's state matters more than the content in front of them, and if that state can actually be measured. Three methods tested those two, existing services and competitors for what the technology can already read, interviews for where focus breaks and why. Each had to end in a flow or interface decision instead of a finding.
Existing Service Analysis
Examined how emerging technologies support cognitive enhancement and learning adaptation.
User Interview
Explored study habits, motivation triggers, stress points, and perceptions of focus.
Competitive Analysis
Examined how emerging technologies support cognitive enhancement and learning adaptation.
research HIGHLIGHTS
Existing Service Analysis highlights
Personalization is the standard claim in online learning, so the first thing to test was whether content is actually the constraint. It is not. Personalization already improves what learners take away, and those same learners still say their condition derails them and their habits do not hold. So the gap is not what gets taught. It is the state the learner is in when it gets taught.
Insight Learners differ in energy and circumstance far more than in content level, so adaptation has to track state and not just skill.
" For every learner on their educational journey "
What personalization already solves
What it leaves untouched
Improve technical skills, self-regulation and resilience with personalized online learning
Say their situation or physical condition affects their learning
Struggle to keep online study consistent
One dot = one learner in a hundred
Insight
Learners differ in energy and circumstance far more than in content level, so adaptation has to track state and not just skill.
User interview highlights
The numbers showed that state matters. They did not show where a session breaks.
Four conversations traced that, three learners aged 20 to 23 and one expert in cognitive measurement. Each broke at a different point, and together they set the boundary for what the system should sense.
Seoyeon | 20 | 3 years in foreign language high school in Korea, 2 years at a university in USA
“Visual content helps me stay focused and understand ideas faster than text-based learning.”
Problem
Text- and lecture-heavy platforms increase cognitive load and encourage memorization rather than understanding.
Insight
Mixed media (video, visuals, interaction) improves engagement, comprehension, and focus—especially for visual learners.
Chaeyoung | 23 | 1 year as an exchange student
“Learning keeps my focus when it starts from what I’m curious about and gives immediate examples when I’m stuck.”
Problem
Existing learning services are not designed around the learner’s interests or real-life needs, and they don’t give immediate examples and corrections when learners get stuck, so focus drops and progress slows.
Insight
Learning experiences should start from learner curiosity, lower entry barriers with short, accessible content, and offer immediate support that connects learning to real-life practice.
Yeonoh | 22 | 3 years at a science high school for gifted students, 6 months as an exchange student
“Consistency matters most to me, but I want learning to adapt to my focus level and correct my mistakes right away.”
Problem
Existing learning services don’t adapt to a learner’s focus level and rarely turn what learners write or say into immediate feedback and repeat practice, making consistency hard to sustain.
Insight
Learning systems should adapt to focus in real time, provide instant correction, and support lightweight, repeatable practice that makes consistency easier.
Professor Youngkyung Park | Director, Color Design Research Institute at Ewha Womans University
"Instead of treating 'attention as a single number, we should distinguish immersion from simple focus and explore whether it can be inferred with feasible signals, then use that state to personalize learning time and content in a technically realistic way."
Problem
Many learning services talk about personalization, but they lack a realistic technical plan for what signals to measure, how to infer learner state, and how to connect that to the system.
Insight
A learning service should target immersion, acknowledge practical limits of EEG, and combine more feasible signals like heart rate with learning logs to infer state, then recommend content format and timing accordingly.
Competitive Analysis Highlights
Competitors treat personalization as a matching problem, pairing a learner with a course and stopping there.
None of them read what happens once the session starts, so nothing changes when attention drops mid-lesson.
That leaves the session itself unclaimed, which is the position Redbio takes.
research summary
Online learning does not fail from a shortage of content. It fails when a rigid structure meets a fluctuating learner. Three breakdowns repeated across the research, habits that will not hold, drop-off from static content, and focus lost to fatigue.
Each maps to one move, a tailored curriculum, curation from bio-signals, and pacing that adjusts while the session runs.
Pain point
Difficulty sustaining consistent learning
High drop-off due to static content
Reduced focus from mental fatigue
Opportunity
Deliver a tailored curriculum
Curate content using bio-signals
Adjust pacing in real time
03 | Define
Insight to Structure
overall project direction
Reading a learner's state is only useful if showing it does not cost the learner attention. So the product splits in two, a dashboard that orients before the session and a learning flow that reveals bio-data only when it changes a decision. That split was a hypothesis at this point, and usability testing is where it got settled.
Persona & Journey
Define Users
Map goals and context
User Flow
Structure Flow
Translate insight into steps
Usability Testing
Validate & Refine
Test, learn, and iterate
persona & journey map development
Based on the research insights, I developed a persona and journey map to map out key moments in the learning flow.
Ethan
16 years old
High school student
Boston
Ethan is 16, studies for science courses and puts in the hours without seeing what they return. His journey breaks in the middle, where the effort feels wasted and he has no evidence of why.
Redbio enters exactly there, replacing a guess about what went wrong with a record of it.
Phase
of
journey
Studying
with Frustration
Discovering
Redbio
Understanding
& Adjusting
Reinforced &
Sustainable Learning
Emotion& Thought
“I’m studying a lot, but it feels inefficient and exhausting.”
“Maybe the problem isn’t effort. Maybe I need to see what’s happening.”
“This feels like it’s teaching me in a way that fits how I’m learning now.”
“Studying feels more manageable when it adapts to me.”
Touch-point
Study
environment
Learning
materials
Redbio
overview
Study-with-
tracking
setup
Adaptive
learning
experience
Pattern
-based
guidance
Ongoing
adaptive
learning
Progress
over time
user flow
The flow had to be technically real, because a state-aware system fails the moment it claims a reading it cannot take. Working with a biosignals expert, I cut every branch that depended on a signal too unstable to trust and let heart rate and learning logs carry those decisions. What survived is a loop that measures, infers and intervenes without asking the learner to confirm anything.
Wireframe
The unresolved question was still how much of the data to surface, and when. Wireframes settled the hierarchy before any visual decision, so the argument stayed about structure instead of style. What they could not settle was whether seeing the data helps at all, which is what the test was for.
Usability testing
The direction assumed learners would want their state visible while they studied. Nobody had tested that.
So the sessions were not scored on task completion. They were scored on whether seeing bio-data helped concentration or competed with it.
- Goals
- Validate the target, test the flow end to end, and find what breaks concentration during learning
- Participants
- 5 university students, aged 20 to 25
- Format
- Moderated, 20 to 30 minutes each. The full flow, then a post-task interview
- Feedback
- Qualitative, on bio-data visibility, its effect on concentration, and dashboard preference
It competed. All five asked for the data, and all five looked away from the lesson each time it moved.
The problem was not the data. It was the motion.
Insight
Users wanted to track their status, but real-time bio-data fluctuations distracted them during learning. We moved detailed live metrics to a secondary hamburger menu and adopted a summary-style dashboard to balance focus and information.
04 | solution
SYSTEM MECHANISM
The system runs one loop and repeats it for as long as the session lasts. It reads heart rate, brainwave and camera signals, infers immersion, continuity and fatigue, then acts on what it found. The action is a content change, an article, radio, video or mixed media, chosen for the state rather than the syllabus order.
Brain Wave
Heartbeat
Face Recognition
Camera
Immersion
Continuity
Tiredness
Redbio Interface
Article
Radio
Video
Mixed Media
Step 1 Data Acquisition
Retrieving real-time biometric data
during learning
Step 2 State Analysis
Measuring immersion, persistence, and fatigue levels
Step 3 Adaptive Intervention
Recommending personalized content for efficient learning
Bringing Strategy to Life
Step 1
Landing Page
Biometric learning asks for trust before it asks for a signup, and hardware raises that bar further. The page shows the all-in-one package, sensors and software, as one concrete object instead of a claim. It ends in a single action, the Redbio Test, so curiosity converts before it cools.
The prototype video may take some time to load. Please wait a moment.
Step 2
BIOMETRIC SENSOR INTEGRATION
Every personalization system opens with a setup screen, which is the first place learners give up. Redbio calibrates through the webcam and sensors in the background, reading stress, fatigue and anxiety without a form. The baseline is set before the learner notices there was a step.
The prototype video may take some time to load. Please wait a moment.
Step 3
PERSONALIZED DASHBOARD
A learner opening the app has already decided to study but not what to study. The dashboard answers that from the current reading and their interests, and it shows state as a summary rather than a live feed. It makes the first choice of the session, which is the choice most likely to be made badly.
The prototype video may take some time to load. Please wait a moment.
Step 4
ADAPTIVE LEARNING INTERFACE
State changes while a session runs, and a fixed lesson has no answer for that. When fatigue rises or immersion drops, Redbio lowers content difficulty or adjusts the interface itself, with no metric on screen. The correction happens inside the session, so a bad twenty minutes does not cost the whole hour.
The prototype video may take some time to load. Please wait a moment.
Step 5
BIOMETRIC PERFORMANCE REVIEW
The session ends and the learner still cannot see what happened inside it. This is where everything held back during the lesson arrives, immersion, fatigue and persistence plotted across the session timeline. Familiar shapes carry it, activity rings and a battery level rather than waveforms, so the reading lands in a second.
The prototype video may take some time to load. Please wait a moment.
Prototype in Motion
A short walkthrough of the prototype in use, highlighting key interactions and how the system responds to user context.
The prototype video may take some time to load. Please wait a moment.
05 | VALIDATION
What i learned
01
"Less is More"
Prioritizing User Goals Over Technology
I led with the sensors, assuming visible real-time graphs would prove the system was working. Testing showed the opposite, that the graphs proved it at the cost of the task they were measuring. Moving them into a post-session summary taught me that technology earns trust by staying out of the way, not by being seen.
02
"Translator of Data"
Making the Invisible Visible through Metaphors
Raw waveforms and numbers meant nothing to participants, who could read the chart and still not know how they had done. Familiar metaphors closed that gap, activity rings and battery indicators in place of signal plots. The job was never to display the data. It was to translate it into something a learner can act on in a second.
03
"From Hypothesis to Validation"
The Power of Evidence-Based Iteration
Redbio ran as hardware and a companion app at a live graduation exhibition, in front of strangers with no one to explain it. Problems that never surfaced in testing appeared in the first minute, almost all of them in onboarding and first use. Rebuilding the flow to explain itself is what convinced me a design is only validated in the conditions it will actually meet.
