Case Study · 2024 — 2025 · Mobile · AI
Product A persistent AI language companion that turns daily practice into a habit.
A Pocket Companion.
9:41

Hello, Emily 👋

Search…
Practice with AI BuddyView all

Chat

with AI
Buddy
😎
AI Buddy
Steven

Talk

with AI
Buddy

Game

with AI
Buddy
Chat HistoryView all
Improve Speaking⋮
Improve Writing⋮
Home
Plans
Explore
Profile

Business Trip
Guidance

Total: 10 Sessions 10 hours
"Short phrases to get close with…

Session List

35%
Session 3
Greet your boss
Session 4
Greet your coworkers
Session 5
Greet your business partners
Session 6
Greet your business partners
Emily, starts your
exploration!
Search…
Resources

Library

Podcast

Translator

Grammar

Explore Study Plan
10 Sessions · 10h 280 Collected

Business Trip Guidance

Published by Natasha
Home
Plans
Explore
Profile
Language That Sticks.

Languru pairs learners with a persistent AI companion — named and personalized during onboarding — who adapts to their pace, picks up where they left off, and turns the intimidating work of mastering a new language into small, daily rituals.

Client
Private individual — wanted to build an AI language tutor as the category took off in 2024, and brought us in to design it.
Role
Product Designer
Product strategy, UX, IA & prototyping
Team
Penghua Zhou — Product Design
Yaxin Cao — UI & Branding
Landscape Designer, SWA Group
Timeline
16 weeks
Jun 2024 — Oct 2024
Deliverables
iOS app design, design system,
onboarding flows, brand
Recognition
Honorable Mention
IDA 2024 — Mobile App Design
The bet
A daily
habit.
A patient companion that picks up where you left off should turn practice into a daily ritual. If it shipped, the metric I'd own is 7-day practice retention — returning to talk, not keeping a streak.
The ask
Beyond
duolingo.
Design a language app that goes past gamified drills — one that actually prepares you for a real conversation in a foreign country.
The insight
Practice,
not points.
Learners stall when lessons feel disconnected from life. What they want is a patient partner to rehearse with — on demand.
The solution
A pocket
companion.
Chat, talk, play. Three modes of AI practice woven around a personalized study plan and a library that grows with you.
01 · Problem & Research

From eight conversations
to three product themes.

01 · Recruit

Screener survey

A short intake survey filtered for adult learners with prior language-app experience and an active reason to keep learning — travel, work, family, or self-directed study. The screener captured current app, target language, and self-rated level to seed segmentation.

Output
Qualified pool
02 · Probe

Semi-structured interviews

Eight qualified participants sat for one-on-one conversations. An eight-question protocol probed daily habits, Duolingo experience, pain points, AI expectations, motivation, and vocabulary strategies — open-ended, with room to follow tangents.

Sample
8 interviews
Format
1:1, semi-structured
03 · Synthesize

Affinity mapping

Roughly 64 observations were transcribed onto sticky notes and clustered by behavior and motivation — not demographics. Three themes surfaced, each pointing toward a distinct product capability that became the foundation of the design direction.

Inputs
~64 notes
Clusters
3 themes
Three themes from the affinity map

Every participant said a variation of the same sentence: "I just want to actually talk." The threads underneath that wish split into three — each becoming a product capability.

Theme 01
No system

No system to learn within

"Duolingo just randomly teaches words. I'd prefer a more logical way to learn." Participants wanted a path shaped by why they were learning — not a single global curriculum applied to everyone.

Design implication
A personalized study plan, tuned by goal.
Theme 02
No scenarios

No scenarios to rehearse in

Vocabulary arrived detached from the situations participants actually needed it for — a clinic visit, a business trip, a conversation with a spouse's family. The scenarios didn't transfer.

Design implication
Scenario-based AI practice — Chat & Talk.
Theme 03
No personalization

No personalization that adapts

Recognition exercises were too easy; speaking was missing. Participants asked for AI that could adjust to their level, weak spots, and learning behaviors over time — not a static curriculum.

Design implication
AI tuned to learning behaviors.
Voice of the user · three interviewees
01
Interviewee 01
Travel & enjoyment
"Partially to communicate better when I travel, partially because I enjoy the learning process."
02
Interviewee 02
Speaking with patients
"To communicate with more people, especially my patients."
03
Interviewee 03
Re-learner · wants structure
"Duolingo doesn't have a specific learning system. I'd prefer a more logical way to learn."

Learners don't lack apps.
They lack someone to practice with.

The market is crowded with apps that teach learners to recognize a word. The moment someone has to produce language in a real situation — order a coffee, run a meeting, comfort a patient — most of them freeze.

What they wanted wasn't another curriculum. It was a safe place to rehearse — with someone endlessly patient.

02 · Process

From fuzzy idea to
validated concept.

01
Week 1 — 3

Discover

8 semi-structured interviews with adult learners; deep competitive teardown of Duolingo and reference scans of TALKIE and emerging AI-language apps.

02
Week 4 — 6

Define

Synthesized findings into personas, journey maps, and three jobs-to-be-done: chat, talk, play.

03
Week 7 — 10

Design

Low-fi wireframes through four iterations, usability-tested at each. Settled on a four-tab IA.

04
Week 11 — 14

Deliver

Hi-fi screens and an interactive prototype in Figma; design system and brand direction finalized with Yaxin.

05
Week 15 — 16

Iterate

A final validation pass surfaced friction; prioritized six high-impact fixes to onboarding and the practice loop.

03 · Architecture

Four tabs. Forty-seven screens.
One mental model.

L Languru — Four tab app
Home 12
Greeting · dashboard
Chat with AI Buddy
Talk with AI Buddy
Game with AI Buddy
Chat History
Search
Plans 11
Pinned plan
Course progress
Session list
Daily quiz
Plan settings
Language selection
Explore 14
Library
Podcast
Translator
Grammar
Community plans
Resource detail
Profile 10
Learning statistics
Ongoing courses
Finished courses
Account & settings
Subscription
Bookmarks
04 · Design

Four rules that
shaped every decision.

01

Warm over clinical

The AI has a name, a face, and a voice. Learners aren't talking to a system — they're talking to Steven, who happens to know nine languages.

02

Context, not content

Every study plan is anchored to a real-life context the learner named during onboarding. "Business Trip to Germany" beats "Unit 4: Travel Phrases."

03

Low-stakes practice

No streaks, no public rankings. Progress is a quiet private ring — visible when you want it, invisible when you don't.

04

One clear next step

At every screen the learner knows exactly what to tap next. No dashboards-of-dashboards, no decision paralysis.

A brand built on warmth,
not tech.

Palette

4 primaries · 4 neutrals
Yellow#FEDE65
Lilac#9D94FF
Blue#587BF3
Ink#000000
Bone#FFFFFA
Rule#D6D3D1
Tertiary#44403C
Muted#A7A29F

Typography

Display · UI
Hallo, Emily
Display
Nunito Sans 900
Italic accents
Fraunces 400 italic
UI / Body
Nunito Sans 400 – 700

The product,
screen by screen.

Home · Daily anchor · Live

A calm dashboard that asks one question: shall we practice?

The home screen doesn't overwhelm. Three big modes — Chat, Talk, Game — and a quiet list of what Emily was doing yesterday.

  • Named AI buddy personalized during onboarding
  • Last 2 conversations surfaced, the rest tucked into history
  • Search as an escape hatch — never the primary path
9:41

Hello, Emily 👋

Emily
Search…
Practice with AI BuddyView all

Chat

with AI
Buddy
😎
AI Buddy
Steven

Talk

with AI
Buddy

Game

with AI
Buddy
Chat HistoryView all
Improve Speaking⋮
Improve Writing⋮
Home
Plans
Explore
Profile
Study Plan · Context-first · Live

Every plan is a real-life reason — not a textbook chapter.

"Business Trip Guidance" is the plan, not "Unit 4." Learners pinned the plan they cared about most, tapped into bite-size sessions, and ended each day on a low-stakes quiz.

  • Plans framed around real-world scenarios from onboarding
  • One plan pinned, others collapsed into a count
  • Add a new language → re-enters the 5-step onboarding
9:41
Hi Emily,
start your German journey
Pinned Study Plan
10 Sessions

Business Trip
Guidance

Viewed 2 days ago
35%
Dashboard
7h
Study Time
Ongoing
Study Plans
5
Home
Plans
Explore
Profile
Explore · Breadth on demand

When the buddy isn't enough, reach for the library.

Some days learners want to curl up with a book or a podcast. Explore is the room that holds all of it — resources, translation tools, and community-published study plans — but only appears when the learner goes looking.

  • Four resources — Library, Podcast, Translator, Grammar — one tap away
  • Community study plans sorted by context, not level
  • Tools (translator, grammar) live here, not in the nav — they're utilities, not destinations
9:41
Emily, starts your
exploration!
Search…
Resources

Library

Podcast

Translator

Grammar

Explore Study Plan
10 Sessions · 10h 280 Collected

Business Trip Guidance

Published by Natasha
Home
Plans
Explore
Profile
Profile · The learner's own room

Progress worth looking back on — not points to chase.

Profile is where a learner's story lives: pinned plans, the week's study time, and the habit streaks that quietly build. No gamified leaderboard, no public scoreboard — just the learner, their avatar, and a calm view of what they've done.

  • Study Plans at a glance, tappable to jump back in
  • Weekly study-time chart — choose Week, Month, Year, or All
  • Hamburger opens My Account: bookmarks, settings, subscription
9:41
Emily avatar
Study Plans
10 Sessions

Business Trip
Guidance

Viewed 2 days ago
35%
Learning Statistic
Total Study Time: 10 hours
Course Completed: 3
15m
MonTueWedThuFriSatSun
Home
Plans
Explore
Profile
05 · Reflection

What I'd do differently
next time.

Build the AI workflow first

This project shipped in late 2024, before AI agents went mainstream. We leaned heavily on generative AI for UI assets, but the workflow was scattered — we hopped between platforms without a shared system. Today I'd treat AI as a workflow problem first, building an opinionated agent stack into design ops rather than reaching for whatever tool was open in another tab.

Front-load in-person time

My partner brought deep landscape and branding instincts, but we worked almost entirely remote. For defined work — asset reviews, component handoffs — async was fine. For ambiguous early decisions like voice, naming, and scenario picks, ideas slipped past deadlines because we couldn't whiteboard together. Next time, I'd front-load a week of in-person time for the fuzziest calls.

Lock the voice partner early

We put ElevenLabs in front of the client as the leading voice partner; she opted to consider it, and the layer was never implemented. For a voice-first product, the partner choice is itself a product decision — not a deferred integration. Next pass, I'd lock it during design so the flows are shaped by its real constraints.

Thanks.

Thanks for reading. Languru taught me that a warm voice goes further than any clever feature.

Back to top
← Previous Lucens AI-MathSolver Next → AR Mixtory
© 2025 · Languru Case Study
Penghua Zhou · Product Design