TikTok RD Assistant
An efficiency-focused AI platform for TikTok’s development team, to streamline knowledge access and daily workflows.
UX Design Intern
AI Design, Visual Design, User Research
UX Designers,
Product Managers,
Software Engineers
2024/11-2025/1
2 months
What I have done
Answer Page
A structured, source-backed answer flow with quick navigation.
Answer Page
A structured, source-backed answer flow with quick navigation.
Design Goal
Create a unified AI that consolidates internal research knowledge, making insights proactively accessible, understandable, and trustworthy?
Passive and narrow usage
AI only supported generation tasks (reports or surveys) and couldn’t proactively answer research questions.
Unclear source transparency
AI responses didn’t show their sources, reducing user trust in the outputs.
Isolated AI capabilities
AI functions were tied to specific modules and couldn’t access across all internal documents.
Target Users
Collaborated with PM, I set 240 surveys and 10 interviews to better understand our user goals
PMs & Researchers
Age: 20–40 years old, predominantly male.
Traits: Tech-driven, analytical, efficiency-focused.
Goal 01
Quickly query past research documents
31% of users struggled to quickly query past research documents and access relevant insights across internal projects.
Goal 02
Extract actionable insights confidently
22% rarely relied on AI answers due to unclear sources, highlighting a trust gap in the experience.
Competitive Research
I also conduct a research including 4 AI tools to understand how AI can guide users, establish trust, and integrate into workflows.
Insights:
Feature 01: Dynamic Entry to Highlight the New Q&A
A dynamic icon distinguishes the new entry, with hover revealing a brief explanation. Also I designed a top banner to promotes the feature to increase visibility and click-through rates.
After atheteam review, I realized the new AI Q&A function was fundamentally different from existing features, creating an architecture conflict. So I designed this distinct entry that opens to a separate page, keeping it set apart from the current structure while highlighting its uniqueness.
Feature 02: Recommendation Cards to Guide AI Exploration
Recommendation cards surface suggested questions upfront, helping users quickly explore AI capabilities, structure their research, and reduce cognitive load when starting a survey.
Keeps the input frame at the bottom for consistency.
Shifts visual focus to the recommendation cards, weakening the input frame’s hierarchy.
Larger card exposure makes them easier to notice.
Less engaging than the collapsed version and less memorable for new users.
Feature 03: Mode-Specific Guidance for Effective Q&A
By providing different modes, the input frame guides users to ask questions in the right context, and makes AI proactively usable across different research scenarios.
Feature 04: Clear Separation of Sources to Build Trust
By separating internal (Lark) and external sources, the feature increases transparency, builds confidence in AI responses, and helps users decide which insights are more reliable.
Driving Execution
When vendor constraints affected development quality, I led QA, creating annotated comparisons and collaborating with engineering to ensure design intent.
This was my first close collaboration with SDEs, where I learned that simple design details can be technically complex. It reinforced the importance of engineering alignment and rigorous QA to deliver a polished UI.
Impact
Being a Proactive Learner
It's common to design for a product you've never used, or a field you're not familiar with. Always stay curious and proactive in learning to prepare for new knowledge.
It's common to communicate with people who think differently than you do. Always listen carefully to understand, and speak boldly to contribute your perspective.
↓ Photos with my amazing team — grateful for them all.
