Let’s collaborate — open to all kinds of opportunities.
Let’s collaborate — open to all kinds of opportunities.
Let’s collaborate — open to all kinds of opportunities.
My Role
Product Designer
Platform
Android, iOS
Revamping AI Guru's Onboarding and Interface to Increase Retention
At PhysicsWallah’s AI/ML pod, I led the revamp of AI Guru’s onboarding flow and chat interface. AI Guru is a personal learning companion chatbot that solves academic doubts, processes image-based queries, and delivers curated notes and videos from PW’s repository. It currently serves over 4 lakh students monthly and handles more than 2 million queries, making it one of the most used products in the ecosystem.
We designed a new Avatar for this Chatbot (with the help of external Visual design partners, Liquidink), one that feels more relatable. With that, there was we found the opportunity to change the visual theme as well and also an opportunity to increase retention.
This revamp wasn’t about building from scratch—it was about refining the experience to increase feature adoption and leverage the aesthetic usability effect to improve retention.
At PhysicsWallah’s AI/ML pod, I led the revamp of AI Guru’s onboarding flow and chat interface. AI Guru is a personal learning companion chatbot that solves academic doubts, processes image-based queries, and delivers curated notes and videos from PW’s repository. It currently serves over 4 lakh students monthly and handles more than 2 million queries, making it one of the most used products in the ecosystem.
We designed a new Avatar for this Chatbot (with the help of external Visual design partners, Liquidink), one that feels more relatable. With that, there was we found the opportunity to change the visual theme as well and also an opportunity to increase retention.
This revamp wasn’t about building from scratch—it was about refining the experience to increase feature adoption and leverage the aesthetic usability effect to improve retention.
At PhysicsWallah’s AI/ML pod, I led the revamp of AI Guru’s onboarding flow and chat interface. AI Guru is a personal learning companion chatbot that solves academic doubts, processes image-based queries, and delivers curated notes and videos from PW’s repository. It currently serves over 4 lakh students monthly and handles more than 2 million queries, making it one of the most used products in the ecosystem.
We designed a new Avatar for this Chatbot (with the help of external Visual design partners, Liquidink), one that feels more relatable. With that, there was we found the opportunity to change the visual theme as well and also an opportunity to increase retention.
This revamp wasn’t about building from scratch—it was about refining the experience to increase feature adoption and leverage the aesthetic usability effect to improve retention.
The Impact
1% uplift in retention, though statistically small, offered deep insights into our assumptions
+10% adoption in image-based doubts and notes feature post-onboarding revamp. A significant jump given the passive nature of the nudge.
Validated UI improvements in community forums before launch—students preferred the new look
Reduced app bloat by optimizing assets (custom 220KB background vs old 900KB one) and earned dev appreciation
1% uplift in retention, though statistically small, offered deep insights into our assumptions
+10% adoption in image-based doubts and notes feature post-onboarding revamp. A significant jump given the passive nature of the nudge.
Validated UI improvements in community forums before launch—students preferred the new look
Reduced app bloat by optimizing assets (custom 220KB background vs old 900KB one) and earned dev appreciation
1% uplift in retention, though statistically small, offered deep insights into our assumptions
+10% adoption in image-based doubts and notes feature post-onboarding revamp. A significant jump given the passive nature of the nudge.
Validated UI improvements in community forums before launch—students preferred the new look
Reduced app bloat by optimizing assets (custom 220KB background vs old 900KB one) and earned dev appreciation
Problem Statement
The old onboarding didn’t clearly explain AI Guru’s full capabilities:
Focused on generic value props, not actionable features
Only ~40% of users discovered the image-based doubt feature
<5% knew about or used the notes feature (it was hidden)
The chat interface had its own issues:
Bland visuals, lack of visual hierarchy in light mode (white-on-offwhite)
Poor contrast and low aesthetic appeal
Constraints:
Had to stay within asset sizing limits
Couldn’t alter the core UI due to functional dependencies—focus was strictly visual and discoverability-driven
The old onboarding didn’t clearly explain AI Guru’s full capabilities:
Focused on generic value props, not actionable features
Only ~40% of users discovered the image-based doubt feature
<5% knew about or used the notes feature (it was hidden)
The chat interface had its own issues:
Bland visuals, lack of visual hierarchy in light mode (white-on-offwhite)
Poor contrast and low aesthetic appeal
Constraints:
Had to stay within asset sizing limits
Couldn’t alter the core UI due to functional dependencies—focus was strictly visual and discoverability-driven
The old onboarding didn’t clearly explain AI Guru’s full capabilities:
Focused on generic value props, not actionable features
Only ~40% of users discovered the image-based doubt feature
<5% knew about or used the notes feature (it was hidden)
The chat interface had its own issues:
Bland visuals, lack of visual hierarchy in light mode (white-on-offwhite)
Poor contrast and low aesthetic appeal
Constraints:
Had to stay within asset sizing limits
Couldn’t alter the core UI due to functional dependencies—focus was strictly visual and discoverability-driven
Our Solution
We overhauled onboarding and subtly upgraded the chat interface.
What Changed:
Replaced symbolic onboarding with stylized illustrations that clearly showed how to use features like image doubts and notes
Revamped chat UI with better colors, spacing, icons, and a custom academic doodle background to fix elevation and visual issues
Used onboarding as the primary nudge to improve discoverability; removed tutorial cards after confirming onboarding alone was effective
How We Did It:
Went high-fidelity from the start, exploring visual inspirations even outside edtech
Validated UI improvements with students in community forums before launch
Collaborated with engineers to optimize performance and reduce bloat
What We Didn’t Ship:
Considered feature tooltips and tutorials but dropped them to let users discover naturally via onboarding (users validated this via posts)
We overhauled onboarding and subtly upgraded the chat interface.
What Changed:
Replaced symbolic onboarding with stylized illustrations that clearly showed how to use features like image doubts and notes
Revamped chat UI with better colors, spacing, icons, and a custom academic doodle background to fix elevation and visual issues
Used onboarding as the primary nudge to improve discoverability; removed tutorial cards after confirming onboarding alone was effective
How We Did It:
Went high-fidelity from the start, exploring visual inspirations even outside edtech
Validated UI improvements with students in community forums before launch
Collaborated with engineers to optimize performance and reduce bloat
What We Didn’t Ship:
Considered feature tooltips and tutorials but dropped them to let users discover naturally via onboarding (users validated this via posts)
We overhauled onboarding and subtly upgraded the chat interface.
What Changed:
Replaced symbolic onboarding with stylized illustrations that clearly showed how to use features like image doubts and notes
Revamped chat UI with better colors, spacing, icons, and a custom academic doodle background to fix elevation and visual issues
Used onboarding as the primary nudge to improve discoverability; removed tutorial cards after confirming onboarding alone was effective
How We Did It:
Went high-fidelity from the start, exploring visual inspirations even outside edtech
Validated UI improvements with students in community forums before launch
Collaborated with engineers to optimize performance and reduce bloat
What We Didn’t Ship:
Considered feature tooltips and tutorials but dropped them to let users discover naturally via onboarding (users validated this via posts)
New Avatar of AI Guru
A new avatar full of new hope and capibilities for the students!
New Avatar of AI Guru
A new avatar full of new hope and capibilities for the students!
New Avatar of AI Guru
A new avatar full of new hope and capibilities for the students!
Less symbols, more clarity
Old onboarding hinted. New onboarding showed—features illustrated clearly to drive adoption.
Less symbols, more clarity
Old onboarding hinted. New onboarding showed—features illustrated clearly to drive adoption.
Less symbols, more clarity
Old onboarding hinted. New onboarding showed—features illustrated clearly to drive adoption.
Smart help, straight from the start (Light Mode)
Stylized walkthroughs highlight key AI Guru features—image doubts, notes, and more.
Smart help, straight from the start (Light Mode)
Stylized walkthroughs highlight key AI Guru features—image doubts, notes, and more.
Smart help, straight from the start (Light Mode)
Stylized walkthroughs highlight key AI Guru features—image doubts, notes, and more.
From bland to brand-ready (Dark Mode)
Cleaner colors, fixed elevation bugs, and better contrast—without changing functionality.
From bland to brand-ready (Dark Mode)
Cleaner colors, fixed elevation bugs, and better contrast—without changing functionality.
From bland to brand-ready (Dark Mode)
Cleaner colors, fixed elevation bugs, and better contrast—without changing functionality.
AI Guru is Thinking...
Built this delightful interaction to turn the time AI Guru takes to think into something interesting.
AI Guru is Thinking...
Built this delightful interaction to turn the time AI Guru takes to think into something interesting.
AI Guru is Thinking...
Built this delightful interaction to turn the time AI Guru takes to think into something interesting.
More screens
See how various scenerios and cases look after changing the theme.
More screens
See how various scenerios and cases look after changing the theme.
More screens
See how various scenerios and cases look after changing the theme.
Explored it all—then trimmed the fat
Explored options to nudge users to explore features—scraped the Idea and decided to rely on intuitiveness only
Explored it all—then trimmed the fat
Explored options to nudge users to explore features—scraped the Idea and decided to rely on intuitiveness only
Explored it all—then trimmed the fat
Explored options to nudge users to explore features—scraped the Idea and decided to rely on intuitiveness only
Reflection & Learnings
My biggest learning? A data-backed hypothesis is still a hypothesis. We assumed introducing features would boost retention, but the reality was more nuanced. Curious students discover more, not the other way around.
That said, the project was a hit. Aesthetic upgrades landed well, adoption improved, and I learned to trust real user behavior over assumptions. Next time, I’ll treat even strong data with a little more skepticism—and a lot more testing.
Thanks for reading :) If you're interested in knowing more about this, DM me!
My biggest learning? A data-backed hypothesis is still a hypothesis. We assumed introducing features would boost retention, but the reality was more nuanced. Curious students discover more, not the other way around.
That said, the project was a hit. Aesthetic upgrades landed well, adoption improved, and I learned to trust real user behavior over assumptions. Next time, I’ll treat even strong data with a little more skepticism—and a lot more testing.
Thanks for reading :) If you're interested in knowing more about this, DM me!
My biggest learning? A data-backed hypothesis is still a hypothesis. We assumed introducing features would boost retention, but the reality was more nuanced. Curious students discover more, not the other way around.
That said, the project was a hit. Aesthetic upgrades landed well, adoption improved, and I learned to trust real user behavior over assumptions. Next time, I’ll treat even strong data with a little more skepticism—and a lot more testing.
Thanks for reading :) If you're interested in knowing more about this, DM me!