Talkzy
An AI companion app built from scratch and
shipped in 6 weeks for Zupee's 100M+ users
Role
Design & Research
Platform
Android
Timeline
6 weeks
Tools
• Figma
• User Research
Industry
Social and
Entertainemnt
A Little Context First
Zupee's Tier 2 and 3 users were not just playing for money. They were playing for time. When RMG regulations started tightening, leadership spotted an opportunity. AI companion apps on the Play Store were seeing unusually high engagement. The user behaviour was familiar. The market was wide open.
So we decided to build one.
The Brief We Started With
Build an AI girlfriend or boyfriend app for Tier 2 and Tier 3 India. Simple enough on paper.
But before a single wireframe was drawn, I wanted to hear it from the users themselves.
How We Found the Truth
No labs. No forms. Just conversations.
I spent time with 8 users around our building complex in Gurugram. Security guards, valet driver, gardeners and admin staff. I asked them two things. Had they ever heard of AI? And would they use an app to talk to someone after their shift?
Nobody had heard of AI. But everyone had something to say about the second question.
Who we spoke to
🚕
Valet Driver
👱♂️
Admin staff
👮🏽♂️
Security gaurds
What They Actually Said
Nobody said they wanted a girlfriend or boyfriend app. Not one person.
But everyone described the same thing. Someone to talk to in their alone time. Someone who could give them real, practical advice. One user wanted to know how to find a one room set in Gurugram within a budget of ₹8,000. Another just wanted someone to listen after a long shift.
The romantic framing made them uncomfortable. The need underneath it was real.
The Pivot
The research made the decision for us. Nobody wanted a girlfriend or boyfriend. They wanted a companion.
That single shift in framing changed everything.
PROCESS
What We Designed
We started wireframing in December and shipped in January. Six weeks from brief to product.
The experience was built around three things. A persona that felt warm and approachable, not robotic. Conversations that could handle real practical questions, not just small talk. And an onboarding that eased users into talking to an AI for the very first time without making it feel strange.
01
Competitive Analysis
Before drawing a single screen, we studied what was already out there.
Replika 30M downloads, 0.5M MAU built around emotional and relationship interaction. Avatar based with deep personality customisation. Heavily western in tone and context.
Chai 25.7M downloads, 1.7M MAU Fantasy and character based companions. Positioned as unlimited and private. Again, built for a global audience with no real India context.
Both apps were doing well globally. But neither spoke to the Tier 2 and 3 Indian user. No cultural familiarity, no practical utility, no Hinglish, no understanding of their daily reality.
That was the gap. An Indianised AI companion that felt familiar, spoke their language and could actually help with their everyday life.
Replika
30M downloads | 0.5M MAU
Emotional + relationship interaction
Chai
25.7M downloads | 1.7M MAU
Themed characters + chat experiences
02
Wireframe
We tested three monetisation models to find the right balance between accessibility and revenue.
Pay per minute Free daily minutes with a lifetime cap. After that users topped up a dedicated Talkzy wallet. A live timer ran inside every chat session.
Subscription with trial After free minutes expired users were prompted to subscribe with a trial. Once subscribed the timer disappeared. Cancellation lived inside bot lobby settings.
Subscription without trial Same as above but straight to subscription with no trial option.
All three models closed the session after 30 seconds of inactivity if the app was backgrounded or killed.
V2 - Nav bar
V3 - Tab bar
Solving for discovery & navigation
V1 - Banner
Explore Screen
Initial Chat
Chat cont.
Initial Framework
Home
Buy minutes
PG
Final state
Pay Per Minute
Free session ended
Trial
PG
Final state
Subscription with trial
Free session ended
Buy pack
PG
Final state
Subscription without trial
Free session ended
03
Final Visuals
From six weeks of research, wireframes and iteration; here is what Talkzy looked like when it shipped.
Onboarding video
Talkzy homepage
Onboarding
Home
Talk time about to over
Buy pack to view picture
Buy subscription nudge
Chat screen states
First five mins free
Onboarding video
Talkzy homepage
Subscription models
Home
Outcomes
Talkzy shipped in January 2026. For a brand new product built in six weeks the numbers told a solid story.
Average talk time
20 mins per day
D2 retention
19%
Onboarding completion
32%
Users were not just downloading it. They were coming back and spending real time inside it. For a Tier 2 and 3 user who had never interacted with AI before, that was the real win.
Key Takeaways
The brief is never the truth.
Research told us to build a friend, not a girlfriend. Listening saved the product.
Context is everything.
Global apps had millions of users. None of them worked for our user. Bharat needs its own.
Onboarding is an introduction.
These users had never spoken to AI before. The first screen was not a flow, it was a first impression.
Clarity beats speed.
Six weeks. One clear insight. That was enough.
NEXT PROJECT
Scaling Zupee
Crafted with love & care
© 2026 Gurmeet Singh. All rights reserved.