Revamping Pi
Modernizing Pi's navigation and basic feature set
Role
Lead product designer
Disciplines
Design systems
Human interface design
Interaction design
Year
2025
CONTEXT
Pi is a consumer-facing chatbot by Inflection AI. In early 2025, the company shifted focus towards revitalizing Pi.
I led the design for 2025 revamp (Web/iOS/Android) where we fixed potholes around navigation and chat creation and introduced new features like deletion and in-conversation feedback.
Background
I joined Inflection AI in the second iteration of the company (see history of Inflection AI). Pi had been in a stagnant, maintenance mode for about a year when company leadership redirected our focus to revitalizing Pi in the following ways:
“Make Pi more mine, reliable, modern, extensible”
As part of this effort, I spearheaded the revamp to fix several apparent usability problems.
Designs
Navigation cleanup
"Threads" were renamed to "Chats" to align with industry standards. Chats were shifted from the secondary navigation bar to the primary navigation to be more accessible.
We took care to not disrupt existing usage patterns for users who primarily used the Home chat (their long, running main chat) but sought to decrease confusion for new users by increasing the visual distinction between the Home chat and regular chats that were more one-off and topic-based.
The mobile navigation was similarly revised to be consistent with the web experience.
Message & chat deletion
In additional to being highly requested by users, allowing chat deletion was in line with the spirit of empowering users with better privacy controls. We wanted to take an additional step and offer users the ability to delete individual messages within chats, giving them more fine-grained control over what information showed up in their chat history.
In-conversation feedback and report mechanisms
Data showed that users were using the existing "report" feature as a way to try to fine-tune the model's responses for themselves. To de-noise the data going into our Safety Center and to help our model team gather data for fine-tuning, I introduced a reaction bar that allowed users to mark a response as good/bad, separate from reporting a message.
Reflection
While Pi had been ripe for updates, it was crucial that we considered how existing users were using the app and how changes would align with our internal goals. By identifying and initiating these revisions, I helped get our product on a stable footing, addressing complaints for existing users and preparing the product for new updates to come.
