10
RESPONSIBILITIES
PHYSICAL UX DESIGN
PROTOTYPING
RESEARCH
Timeline
March-October 2026
context
Designing the hardware interaction of a connected AI beauty mirror
At Ammunition, I worked as a Physical UX Designer across a range of products. For Ester, I was the only hardware UX designer on this project, working on a category-defining AI beauty mirror from kickoff through manufacturing readiness. I collaborated with industrial designers, Ester’s brand, hardware engineering, and digital teams to prepare the mirror for launch.
My work spanned interaction design, digital and physical prototyping, and usability testing. I focused on how the mirror’s physical controls and light and sound cues worked with the companion app to help people learn and repeat a skin scan at home.

the problem
The gap in skin tracking solution
Ester saw a gap between the skincare decisions people make every day and the feedback they get about what’s working. Phone-based tools make skin analysis accessible, while professional systems like VISIA ( very expensive! ) captures details that ordinary photos can miss. Ester’s goal was to fill the gap in-between.
Phone is too unreliable
Phone-based tools make a skin check easy to start. Changes in lighting, distance, and angle can make photos harder to compare over time.

Visia is too exclusive
VISIA uses controlled lighting to reveal different skin features. It gave us a benchmark for consistent imaging in clinics and at beauty counters.
Design Opportunity
Make a clinical procedure delightful and repeatble for everyday Use
While achieving precise face positioning rigorously like a clinical-grade device is important for data analysis and embodying trust, Ester is also designed to be a luxury beauty object, not a tech gadget. Elevating the daily ritual of skincare and makeup is equally important.
Seamless Face Positioning
Help people find the target distance and angle, then clearly confirm when they’re ready for capture.
Create Ritual
Make the scan brief and easy to repeat alongside the skincare routine people already have.
Assuage Privacy Concerns
Make it easy to see when the camera is physically covered.
The process
Zooming in and out
I explored broadly enough to surface questions the team hadn’t yet asked, then narrowed in on the decisions that mattered most to the scan. I used working prototypes to test our assumptions early. That helped us see where people got confused, where the hardware might limit us, and what we needed to resolve next.
Interaction ideation
Explore what’s possible.
Buidling scenarios
Narrow to promising directions.
Prototyping light & sound
Surface risks early.
Control mapping
Prioritize everyday actions.


User flow
Mapping out the big picture
I mapped what the user does alongside what the mirror needs to sense, decide, and communicate. At each step, I flagged the inputs, outputs, indicators, and controls the interaction might need.
That gave industrial design, strategy, and the client a shared view of the scan. I defined what the mirror had to detect and communicate while leaving room for the team to decide how.bendum urna ipsum, vel rutrum sapien tristique vel.
the goal
How might we:
… guide user to the exact distance and angle relative to the mirror?
A repeatable scan depended on finding a target about six inches from the mirror and turning to the required angles. The guidance needs to help users adjust, recognize alignment, and know when to hold still.



ideation
Exploration around key scanning moments
The mirror serves several functions, but the scan is central to Ester’s promise of helping people track changes in their skin over time. I started there, treating ideation like gathering ingredients before cooking. I explored what physical guides, controls, and feedback could each offer before deciding how to combine them into full scenarios.
scenarios
Narrowing down through building Scenarios
I started with four distinct approaches to scanning. Each would change the mirror’s hardware architecture and redefine the role of the mirror, the phone, and the user.
Hover to see the storyboards!


Physical guidance


Computer vision


Smartphone-assisted


Smartphone-centric


Physical guidance

Computer vision

Hybrid with proximity sensor
the new challenge
Round 2, How might we:







prototype
Exploration through interactive prototyping
I built seven webcam prototypes in p5.js so the team and client could try the concepts in a browser. Internal testing gave us an early read on which light and sound cues made sense as people moved.


limitation
Pilot twist! New hardware limitation unveiled
At the client review, we learned that cost and lighting architecture limited the mirror to two LED panels without individually controllable segments. Surfacing that constraint while we were still exploring helped us quickly pivot.
Despite the early interface concepts relying on a fully addressable light ring, the client found my exploration inspiring. The prototypes showed ways to combine light and sound they hadn’t considered, like using a chime to make finding the right position feel satisfying or using countdown to tell people how long to hold still with their eyes closed. We carried those ideas into the two-panel design to make the scan easier and more enjoyable to follow.
interaction design
Light behaviors design
I approached the LED cues the same way I approached the early concepts. I laid out my ingredients first: pattern, hue, and layout. Then I explored how to combine them to guide someone’s distance from the mirror and the angle of their head.
I asked what a steady light might signal compared with a slow pulse, which hues felt comfortable up close, and where a cue would still be visible when someone was inches from the mirror or turning their head away. Those questions helped me choose which combinations to prototype.
Pattern



Hue
Layout



Audio cue design
I used Ableton and animation to explore ways sound could guide positioning: a single confirmation tone, a pitch that rose as users moved closer to the target, and a chord progression that conveyed the same progress through harmony.
I then narrowed the scan to a small set of essential cues. Sound mattered most around capture, when users had to close their eyes before the flash. A tone asked them to close their eyes and hold still, and a distinct chime confirmed capture and told them they could open their eyes again.
Building these cues taught me how to use pitch, harmony, and timing to communicate what the user needed at each step.
Confirmation

Gradual pitch shift

Chord Progression



LED language
I mapped what could go wrong before, during, and after a scan, from a covered camera or a face moving out of range to a processing failure. A single LED had to make it clear whether to adjust their environment, check the camera cover, open the app, or reset the mirror.
Even with a single LED, I could change its color, blink pattern, and how it transitioned between states. I started by asking what each behavior might remind someone of and how that association might make them feel. A quick pulse can evoke a racing heartbeat and suggest urgency, while a slow fade can recall steady breathing and make a wait feel calmer. I used those associations to decide how each cue should draw attention.
Hardware control mapping
I ranked the mirror’s actions by how often people used them and what could happen if they triggered one by accident. A survey showed that people tended to stick with one hue and adjust brightness more often, so I gave brightness the most direct gesture and made hue switching less prominent. Starting a scan activates the camera, so I wanted that action to take a deliberate gesture.
As for the specific gestures mapping, I borrowed mental model from everyday controls to each concept a familiar starting point. I explored what a turn, press, or double press should do, whether the dial should have tactile detents and how many, and whether a hard stop could mark an off position.
I also helped the product manager source buttons and dials so we could compare turning torque, press resistance, and the feel of detents and hard stops. When I couldn’t find an off-the-shelf part with the tactile response I wanted to test, I modeled and 3D-printed a prototype. I designed the parts to be easy to print and assemble so we could try the feel on physical hardware.
Prototype
A working prototype surfaced engineering risks
I built a working Raspberry Pi prototype to see what would happen at the actual scan distance. The first camera couldn’t fit the whole face in frame. A wide-angle lens solved that problem, but its distortion made tracking harder. I used OpenCV to correct the image, and at one point my setup tracked the face more reliably than the client’s prototype. Building the scan exposed issues I could bring to the client’s engineers early, with a working setup we could compare.

Building a bridges
I defined what each light behavior meant: a slow pulse asked users to move closer, a fast pulse asked them to move back, and a steady light confirmed they were in range. The visual designer could then shape how each cue felt. I built a control panel inspired by the After Effects timeline so they could adjust timing, brightness, and easing in a familiar interface and see the changes directly on the mirror.
Hardware control gestures
Several dial mappings made sense on paper. I built the most promising ones on physical hardware so we could live with them for a bit. Repeating everyday actions and switching between functions let us feel which gestures came naturally and which made us pause to remember what to do.

NEXT STEP
The journey isn’t over
The scan is one part of a larger experience. My work also touched how people first meet the mirror when they open the box, how the app teaches them to scan, and how those first interactions help them build a routine. I’m looking forward to seeing how people make Ester part of their day once it reaches their homes.














