Highlights
ESTÉE LAUDER COMPANIES
Accessibility-First AI Experience
Designed a voice-first mobile experience for users across the vision-impairment spectrum, from low vision to fully blind.
Trust under Uncertainty
Built conversational patterns and recovery flows that kept users confident even when computer
vision was imperfect or ambiguous.
Background
The Goal
Explore how emerging AI technologies could make beauty more accessible to 290 million people globally living with vision impairment.
The Audience
Users across the full vision-impairment spectrum, from mildly to fully blind, each relying trusting feedback from others or non-visual queues.
Role
Senior Product Designer
0-> 1 Pilot
Accessibility-First
Voice-First UX
AI Interaction Design
End-to-end Product Design (Accessibility & AI)
Leading product design and content strategy
Overview
I led end-to-end product and experience design for VIME, covering research, interaction design, conversational UX, voice content strategy, and testing.
Unexpected Reorganization
Our senior content strategist unexpected rolled off , I took ownership of content and voice strategy to maintain continuity across the experience.
Day-to-day
I partnered closely with cross-functional teams and accessibility experts to align user needs with technical constraints and deliver a reliable voice-first mobile experience.
Key Responsibilities
Led accessibility-focus research with visually impaired participants
Designed voice and conversational interaction patterns
Helped to define requirements around computer vision accuracy and responsiveness
Created content and voice principles to guide AI responses
Defined success metrics that balanced user trust with technical constraints
Designing with Non-Visual Constraints
Creating an accessible and delightful experience without a traditional UI
VoiceOver Conflict
Challenge: VoiceOver is essential for iOS navigation, but VIME's voice assistant needed to guide users through makeup application. Running both simultaneously led to audio conflicts, meaning two voices talking over each other.
Design Decision: Rather than fighting VoiceOver or ask users to disable a trusted tool, I reduced reliance on traditional UI interactions, ensuring a cohesive experience that works without VoiceOver interfering. VIME relied almost solely on voice-in / voice-out interaction without gestures or buttons.
Latency Masking
Challenge: Through interviews, I learned that almost all users increased VoiceOver's speed to 3-5x speed. VIME needed at least 3 seconds to analyze and provide feedback. This created a feedback void where users felt like the system was lagging or broken.
Design Decision: I focused on the human experience and natural conversation to mask the delay and create a buffer.
User
Okay, I'm ready!
VIME
Sure, it would be my pleasure to take a look…
By the time that sentence finished, the AI was halfway through processing. This interaction matched the luxury boutique experience, bought time for the system to process, made the interaction feel natural and human.
Natural Lip Color vs Lipstick Color
Challenge: Certain colored lipsticks closely resembled bare lips, making it difficult for computer vision to detect lipstick boundaries.
Design Decision: A conversational fallback that invited the user back into the flow as a built-in re-do opportunity.
User
Okay, I'm ready!
VIME
I'm sorry, I don't detect lipstick. Have you applied it yet? If so, I can take another look.
This helped to build trust through transparency and allowed the AI another chance to re-analyze. Framed as a collaborative effort that brought users into the error handling loop, rather than just presenting an explicit error message.
Positioning, Lighting, and Orientation
Challenge: It's difficult for users to know if their device is properly oriented at their face. One participant said another app just kept yelling “I can’t see you!” over and over again. She was so frustrated, she threw the phone into a drawer.
Design Decision: I explored directional audio, vibration cues, environmental feedback, but interview made it clear that users preferred simple, human instructions.
VIME
I can’t quite see you…try moving your phone to the left.
Research
Focused on understanding vision impairment, non-visual interaction patterns, and how trust is built through transparency while stress-testing AI and computer vision in real time.
Live Pilot (TestFlight)
9 Participants
Accessibility Testing
In-Context Interviews
Understanding Users, Accessibility, and Technology
Learning how users perceive, trust, and recover from AI-driven voice interactions
Technical Constraints and Feasibility
Defined MVP capabilities for computer vision and real-time feedback.
Partnered closely with engineering to understand constraints around:
Detection accuracy
Latency
Spatial awareness
Voice response timing
Understanding Vision Impairment
Consulted with accessibility experts and aligned to WCAG and RNIB standards.
Built empathy through hands-on immersion by navigating an iPhone blindfolded using VoiceOver, developing a deeper understanding of cognitive load, pacing, and emotional reassurance in non-visual experiences.
User Interviews and Testing
I ran a live TestFlight pilot with nine participants with varying levels of vision impairment.
Usability tests were built around intentional failure states, not just a happy path
Real-time troubleshooting of AI mistakes, response latency, camera positioning, and feedback accuracy.
Authenticity and transparency built more trust than a seemingly perfect system
User Feedback
Users valued voice-first interactions without the need to press buttons or use gestrues
Calm, natural, and friendly voice responses increased user confidence, delight and trust
“You know how people say, "I don't know how I ever lived without it"? I think this is going to be one of those apps"
— VIME Testing Participant
Impact
Empowered Independence
VIME validated that voice-first interactions empowered users with vision impairment to confidently and accurately apply lipstick independently, often for the first time.
Trust as a Requirement
The pilot revealed that transparency, conversational pacing, and collaboration mattered more to users than flawless detection, reshaping how success was defined for AI-driven experiences.
