An AI companion that connects online and offline growth journey.
Designed the end-to-end product experience for Pebble — a seed-funded AI emotional companion app based in Shanghai, built on professional psychology frameworks, currently in beta.
About Pebble
A seed-funded mental health solution platform that connects online AI support with offline healing spaces. Driven by AI agents rooted in professional psychological theories, it serves as an accessible entry point, offering users a complete growth journey from online tools to offline experiences.
The Team
Psychology
Yale-trained psychologist advising on the dialogue framework and clinical methodology.
Product
Ex-Alibaba PM with experience scaling consumer products across China's mobile ecosystem.
Business Strategy
10 years of commercial strategy experience across NetEase and Alibaba.
The Users
User 01
The Lightly Stressed
Ages 20–35: working professionals, students, freelancers. Experiencing stress, anxiety, or low motivation — but not at clinical levels. They need fast, accessible emotional support, goal clarity, and actionable guidance. High acceptance of free tools; willing to pay for deeper reports and experiences.
User 02
The Growth Seeker
More detail is proprietary — reach out if you'd like to know more.
User 03
The Clinically Underserved
More detail is proprietary — reach out if you'd like to know more.
Product Thinking
Key Conversion Points
Free → Pro
Build trust, then convert through depth
Build trust through free AI conversations. Convert through demand for deeper Reports, Insight summaries, and structured Courses.
Pro → Max
Position therapy as value, not volume
Position the therapist marketplace as a cost-effective alternative to traditional therapy — the core hook is value, not volume.
Max → One-time purchases
Quality experience drives repeat purchase
Quality experiences build brand loyalty. Repeat purchases follow naturally from users who already trust the product.
Simulated course overview
Insight driven design changes
Key Observations from 15 User Interviews
Observation 01
Emotional triggers shaped the base model's prompt engineering
User interviews revealed relationship stress — especially within East Asian family dynamics — as the primary motivation to open the app. This focused the base model's prompt engineering accordingly.
Observation 02
Demand for personalisation drove IP character design
Users wanted the AI to feel like theirs — distinct, consistent, and recognisable. This directly informed the decision to invest in IP character design.
Observation 03
Turning the trust gap into a monetisation pathway
The gap between AI and human connection shaped the therapist handoff and marketplace feature.
AI Workflow
I used AI not just as a product feature, but as a design tool — accelerating the system-building process itself.
Token Generation
Used Claude to generate a complete design token system from a product brief.
Figma Finetune
Imported tokens into Figma and refined against real screens. AI does the scaffolding; designer does the craft.
Component System
A reusable library covering 25+ screens — one source of truth for design and engineering.
Design System — Version 1
Design
This is an early prototype, since then its been few iterations, please contact JOY for latest designs.
Fig 1: AI voice model fully deployed, caption design for better readability in progress
Next Steps
Currently brewing — AI Memory Adjust Panel
How to transform a PM requirement into clear design directions? Please contact Joy for details.
Voice emotion recognition
Beta users flagged that the AI sometimes misreads tone — responding to "I'm fine" without catching that it was said through tears. The next design challenge is surfacing prosodic signals to the model, so the product responds to how something is said, not just what.