An AI-powered English learning chatbot for non-native speakers, providing personalized support from general proficiency to standardized test prep through conversational AI and scaffolded learning pathways.
Year
2024
Company
Articulate.AI
My Role
Founding Product Designer
Conversation Designer
Target Users
Adult English learners
aiming for skill growth and exam success
Team
CEO
Front-end Engineers
AI & Learning Scientists
Product Designer
0+
0+
registered users within the first week across 100+ countries.
0%
0%
would recommend EduBot to others.
0%
0%
feel more confident in their English skills after using EduBot in conversation practice
0.0/5
0.0/5
usefulness rating based on participant feedback using a 5-point likert scale
01
Grammar-heavy learning
02
One-size-fits-all curricula
03
Limited speaking practice
04
Low real-life exposure
How might we design a flexible, chat centered learning platform that supports both real life English and academic test prep, while making the experience feel conversational, motivating, and personalized?
Literature Review
Competitive Analysis
Field Observation + Semi-Structured Interviews
Personalized Dashboard for users to track their learning progress
Users will receive feedback of the conversation from EduBot at the end of each chat
Users can choose to have a conversation with EduBot based on different topics
Guided conversation to practice integrated language skills and micmic a real classroom session
Random or self-chosen interactive activities that simulate real classroom experiences
Iteration 1 : Linear Sequence
Concept Testing Findings:
01
Practicing all 4 skills in one topic was overwhelming.
02
Topics were engaging, but learners wanted more depth and real-life scenarios.
03
Desire for flexibility to choose or switch skills during a session.
Iteration 2 : Topic Focused Practice Flow
Iteration 2 : Exam Focused Practice Flow
01
The “New Conversation” button was hidden when users scrolled, causing difficulty starting a session.
After
Fixed the button in the left menu bar, making it always visible.
02
Long bot messages combined lesson, hints, and prompts into one, making it hard for learners to parse.
After
Added emojis/icons before each part (lesson, hint, bot question) to visually separate them.
03
Users struggled to follow along when listening to audio and reading at the same time.
After
Added dynamic text highlight as it’s spoken.
How could we design feedback that balances accuracy with the warmth and nuance of human teaching?
01
Tone matters as much as accuracy
Harsh or overly direct corrections can discourage learners.
02
Human teachers mix feedback styles
From explicit correction to gentle guidance and encouragement.
03
Variety sustains engagement
Changing feedback styles helps learners stay engaged.
01
Turning Harsh Red into Gentle Guidance
02
Varying Feedback Like a Teacher
User Input:
“Would you like a window or a aisle seat?”
Different types of corrective feedback EduBot gives:
Types of Feedback
Explicit Correction
Repetition
Recast
Metalinguistic
Elicitation
Meaning
Clearly indicating that the student's utterance was incorrect, the teacher provides the correct form
The teacher repeats the student's error and adjusts intonation to draw student's attention to it
Without directly indicating that the student's utterance was incorrect, the teacher implicitly reformulates the student's error, or provides the correction
Without providing the correct form, the teacher poses questions or provides comments or information related to the formation of the student's utterance
The teacher elicits the correct form by asking or pausing to allow learners to correct their utterances
EduBot Feedback
“Good try! The correct way is: “Would you like a window or an aisle seat?”
“Almost! You said “a aisle seat?” (repeating the error with slight emphasis)
“Would you like a window or an aisle seat?”
“Almost there! Before words starting with a vowel sound like “aisle,” we use “an” instead of “a.”
“Nice try! Let’s say that again: Would you like a window or ...?”
Use Case
Best for beginners who need clear, direct correction to avoid reinforcing mistakes.
Useful for raising learner awareness of what they said, without immediately giving the answer.
Great for fluency practice and lower-pressure contexts, where learners can absorb the correct form naturally.
Effective for intermediate to advanced learners who benefit from grammar explanations and rules.
Encourages active problem-solving and works well for learners who can self-correct with a hint.
01
Leading the Design Process End-to-End
Owning the entire design process from research to prototype taught me to make independent, evidence-based decisions while staying aligned with business goals. Collaborating with the CEO and engineers helped me turn broad ideas into clear, actionable design directions that moved the product forward.
02
Driving Alignment Through Clear Communication
Without other designers on the team, I learned to communicate design decisions effectively across disciplines, using prototypes, conversation maps, and usability findings to align everyone toward the same user-centered direction. This strengthened my ability to lead discussions and make informed design trade-offs.
Through this project, I learned that designing an AI tutor isn’t just about building features, it’s about building trust. Many users wanted to learn with AI, but they still needed warmth, empathy, and guidance. In the future, I’d explore how EduBot’s tone and personality could adapt to different learner emotions to make the experience feel more human and supportive.