A focused empty state avoids a blank dashboard and guides the learner toward the first useful action.
The project itself
Project Overview
Calibrate is an AI-assisted study app built around learning science principles.
The app turns imported notes into source-linked review cards, then helps learners test what they can truly explain instead of simply rereading or recognizing information.
By comparing confidence with answer quality, Calibrate detects false mastery, highlights missing concepts, and turns each review session into a targeted learning plan.
Problem:
Students often confuse familiarity with understanding.
Most study tools help learners repeat or recognize information, but they do not always reveal whether a concept can be explained without support.
Calibrate focuses on active recall, explanation, and confidence calibration to reveal the gap between what learners think they know and what they can actually retrieve.
Goal:
Design a study experience that tests real understanding, not just recognition.
The app should turn notes into source-linked cards, ask learners to rate their confidence before answering, analyze the quality of their explanation, and guide the next session around false confidence and missing concepts.
My role:
Product Designer
I shaped the product concept, defined the core learning loop, designed the mobile experience, and explored how AI feedback could become clear, useful, and grounded in the learner’s own notes.
Responsibilities:
Product framing,
UX strategy,
Cognitive learning research,
User flows,
Low-fidelity wireframes,
AI feedback logic,
Interaction design,
UI, Prototype, iterations, …
All about the user
Research & learning insights
I approached the research from two angles: how learners revise in practice, and what learning science says about durable understanding.
The research focused on active recall, confidence calibration, and the difference between recognizing information and being able to explain it without support.
The main opportunity was to help learners identify the gap between what feels familiar and what they can actually retrieve.
Pain Points
Familiarity feels like mastery
Learners often feel confident when material looks familiar.
Product decision
Calibrate asks learners to explain before showing the source.
Feedback is often too shallow
Most tools show if an answer is right or wrong, but not why.
Product decision
Feedback is linked to highlighted parts of the learner’s own answer.
Review sessions lack prioritization
Learners know some cards were difficult, not what needs attention next.
Product decision
Answers are converted into learning states that guide the next review.
Learner profiles
Instead of building demographic personas, I focused on learning behaviors.
The goal was to understand how different learners revise, where false confidence appears, and what kind of feedback helps them improve.
Competitive audit
I analyzed common study and learning apps to understand how they support repetition, recognition, and engagement.
The audit showed that most tools are good at helping learners review more often, but less effective at revealing whether a learner can explain a concept without support.
Most tools help learners review more often, but few reveal whether they can explain a concept without support.
This gap shaped Calibrate’s core loop: confidence, open recall, explainable feedback, and review priority.
From recognition to calibrated recall
The product logic
The core idea behind Calibrate is that understanding should not be measured by recognition alone.
A learner may recognize a concept and still struggle to explain it without support.
Calibrate turns each review card into a calibration loop: the learner rates their confidence, explains the answer, and receives feedback based on what was actually retrieved.
Appmap
It's a structured scheme that outlines the pages and content hierarchy of the app.
Confidence before answering
Learners rate how ready they feel before seeing feedback, making the confidence signal more honest and less biased by the result.
Explanation quality
The answer is analyzed for covered ideas, missing concepts, vague reasoning, and off-topic parts.
Missing concepts
AI identifies which key ideas were not retrieved and uses them to classify the card into a learning state.
This logic became the foundation of the full learning loop: every answer becomes evidence for what the learner should review next.
Understanding is not measured by recognition. It is measured by the ability to explain without support.
The clear version :
Key UX decisions
After defining the logic, I focused on a few decisions that made the experience both cognitively useful and simple enough for mobile use.
Each choice was designed to reduce passive studying, protect the recall process, and turn AI feedback into a clear next action.
1. Start with import, not onboarding
New users are sent directly to the first useful action: importing study material.
Because Calibrate only becomes valuable once personal notes are added, I avoided a long onboarding or an empty dashboard and focused the first screen on creating the first review set.
Protect recall from visual cues
During the answer step, the interface hides the source, the expected answer, and the card’s learning state.
This keeps the card neutral and prevents learners from linking visual cues to the answer. Some questions can also be slightly reformulated across sessions to check whether the learner understands the concept, rather than memorizing a specific wording.
3. Design for effortful retrieval
Instead of asking learners to select an answer, Calibrate asks them to retrieve and explain the concept from memory.
This creates a useful difficulty: learners may hesitate, make mistakes, or miss concepts, but those moments reveal what actually needs feedback.
Voice mode captures the answer in one pass, training spontaneous oral recall closer to the pressure of an exam situation.
4. Make AI feedback inspectable
AI feedback is not reduced to a single score because a score alone does not explain how to improve.
Learners need to see which parts of their answer were strong, vague, incomplete, or missing.
By linking feedback to highlighted segments of the learner’s own wording, Calibrate turns correction into an explainable diagnosis.
5. Guide the next review, but keep user control
Calibrate recommends a Smart Review based on the current learning states, so the next session is not random.
At the same time, learners can still choose Weak spots only when an exam is close, or Full set when they want to review everything. Session length also lets them adapt the review to their available time without losing the learning priority.
During design exploration, I considered common study situations: quick review, exam preparation, and full-set revision.
Full Flow
End-to-end learning loop
After defining the product logic, I mapped the full learning loop: from imported material to a prioritized next review session.
The goal was to show how each interaction supports the same principle: learners should not only review cards, but understand what they can actually explain.
For a new user, Calibrate starts with the fastest path to value: importing study material.
The first screen avoids a long onboarding or empty dashboard and focuses on one useful action — creating the first review set.
Learners can combine PDFs, photos, and pasted text in the same import. If the material contains distinct topics or courses, Calibrate can organize it into separate review sets.
A focused empty state avoids a blank dashboard and guides the learner toward the first useful action.
The import sheet lets learners add different formats in one place: PDF, photos, or pasted text.
Imported material can be organized into separate review sets when topics differ.
Generate source-linked cards
AI turns imported material into structured, source-linked review cards.
After import, Calibrate makes the AI process visible instead of treating it like a black box.
The app shows that the material is being analyzed, key concepts are extracted, and source references are found.
Before studying begins, learners receive a generated set they can inspect, edit, and trust.
A transparent loading state shows what the AI is doing.
Detected concepts and sections make the output feel structured and verifiable.
Generated cards appear before any learning state is assigned, giving learners control before review starts.
Rate confidence
Learners assess how ready they are to explain before answering.
Before answering, Calibrate asks learners to estimate how ready they feel to explain the concept without looking at their notes.
Capturing confidence before feedback keeps the signal honest: it is not influenced by the result, the source, or the AI diagnosis.
This confidence rating is later compared with answer quality to detect false mastery, underconfidence, or real understanding.
Confidence is framed as readiness to explain, not as a guess about being right or wrong.
The selected confidence remains editable during the answer step in case of a mis-tap.
Answer by voice or text
Learners explain the answer in their own words, either by speaking or typing.
Calibrate asks learners to retrieve and explain the concept instead of choosing from predefined options.
This protects the recall process: the source stays hidden, and the answer must come from memory.
Learners can answer by voice or text depending on their context.
Once selected, the response mode stays active throughout the session to reduce repeated micro-decisions.
Voice mode captures the answer in one pass, training structured recall under mild exam-like pressure.
Typing gives learners more control while keeping the task open-ended and explanation-based.
The hidden source helps Calibrate evaluate actual retrieval instead of assisted recognition.
5.1. AI review
AI turns each open answer into a diagnosis of understanding.
After the learner answers, Calibrate compares the explanation with the expected concepts from the source material.
The goal is not to return a simple right-or-wrong score, but to identify what was clearly retrieved, what remained vague, and which concepts were missing.
The first feedback layer gives learners a quick overview: learning state, short verdict, and highlighted parts of their own explanation.
The review state shows that the answer is being transcribed, compared with the source, and prepared for feedback.
The first feedback view combines a quick verdict with highlighted parts of the learner’s answer.
Specific segments show what was strong, incomplete, or conceptually weak.
5.2. Segment-level feedback
Highlighted segments can be opened to understand specific parts of the answer.
Rather than stopping at a general verdict, Calibrate lets learners inspect specific parts of their answer.
Each highlighted segment opens a focused explanation tied to the learner’s own wording.
This makes feedback more actionable: learners understand what was weak or strong, why it was marked that way, and how to improve the next attempt.
Missing concepts are translated into a concrete idea to retrieve next time.
Weak segments explain why the phrase was incomplete and suggest a focused retry path.
Strong segments can also be inspected to reinforce accurate reasoning.
6. From feedback to review priority
After each session, cards are grouped by learning state so learners know what needs attention next.
Once the session is complete, Calibrate turns each answer diagnosis into a learning state: Blind Spot, Fragile, Solid, or Mastered.
Instead of ending with a generic score, the app shows where understanding is strong, where it is still fragile, and which cards need priority.
The next review is shaped by the learner’s actual answers, while still allowing them to adjust the focus and session length.
Cards are grouped by learning state, making progress and weak areas immediately visible.
The review set keeps learning states after the session, turning diagnosis into a persistent study structure.
Learners can follow Smart Review or adjust the next session based on their current goal.
7. Card detail & learning history
Each card keeps track of what was understood, missed, and should be improved next.
After a review session, each card becomes more than a question: it keeps a learning history.
The detail page combines the reference answer, the latest attempt diagnosis, confidence, answer quality, and covered or missed concepts.
This helps learners understand why a card is Blind Spot, Fragile, Solid, or Mastered — and lets them practice it again immediately.
Learners can reopen any card to understand why it was classified and what needs attention.
The card detail brings together answer, source, diagnosis, and learning history.
From the detail page, learners can practice the card again without waiting for a full session.
Design exploration
Designing the AI review
I explored different ways to make AI feedback useful, understandable, and actionable.
Checklist feedback
This model separated the answer into covered, partial, and missing ideas.
It was precise and easy to audit, but it felt too mechanical. Learners could see what was missing, but not always understand the main learning signal.
Percentage-based scoring
This model translated the answer into a comprehension score, with learned, partial, and missing concepts.
It was easy to scan, but too reductive. A percentage could hide the specific reasoning mistake the learner needed to fix.
Micro-synthesis
This model gave a short AI verdict summarizing the main issue in the learner’s answer.
It felt clearer and more human, but was not enough on its own. Learners still needed evidence from their own response to understand why the verdict was given.
Designing the AI review
Final direction
The selected direction combines a short verdict, highlighted answer segments, and focused explanations.
The verdict gives learners a quick signal, while the highlighted answer makes the diagnosis traceable to their own words.
This makes AI feedback easier to trust, because learners can inspect why a part was marked as strong, vague, incomplete, or missing.
Final reflection
Outcome
What the prototype demonstrates.
Calibrate reframes studying from “Did I recognize the answer?” to “Can I explain it without support?”
The prototype shows how AI can be used beyond card generation: it turns each answer into a diagnosis of understanding, then converts that diagnosis into a clear review priority.
The final concept helps learners:
- identify false confidence,
- understand what was missing,
- revisit weak concepts from their own notes,
- and start the next session with a focused plan.
Key learnings
What this project helped me clarify as a Product Designer.
Calibrate helped me understand that AI feedback should be designed as a learning interface, not just an automated correction
The main challenge was balancing cognitive depth with mobile simplicity: giving enough feedback to be useful without making the experience feel heavy.
It also reinforced the importance of designing the full learning loop — input, recall, diagnosis, prioritization, and next action — rather than isolated screens.