✦ LUMA · PRODUCT DESIGN · 2026

Designed for the event, not the person going to it

Designed for the event, not the person going to it

Luma handles the event. It leaves three moments to chance — knowing what you want, meeting someone there, and staying in touch after.

Luma handles the event. It leaves three moments to chance — knowing what you want, meeting someone there, and staying in touch after.

Role

Interaction design lead —

onboarding, Event Buddy,

post-event flows, design system

Tools

Figma, Figma variables,

FigJam

Team

Christina Valdivia

Naveet Vaid

Minjun Byun

Timeline

Summer 2026

THE SHORT VERSION

Three moments Luma leaves to chance

Three moments Luma leaves to chance

BEFORE

What kind of room am I comfortable in?

A twelve-question preference survey that tells you something about yourself, not just the algorithm.

A twelve-question preference survey that tells you something about yourself, not just the algorithm.

Onboarding · Completion portrait · Event tags

DURING

I’m here alone — who do I talk to?

Pair with another attendee before you arrive. Four respondents preferred a buddy; four were comfortable either way.

Pair with another attendee before you arrive. Four respondents preferred a buddy; four were comfortable either way.

Event Buddy · Plans · Chat

AFTER

That was good. Now what?

Pick up to three people who checked in. Neither of you is told unless you both choose.

Pick up to three people who checked in. Neither of you is told unless you both choose.

Reconnection · Recently Met · Plan Together

Luma is excellent at the event itself. This redesign builds the parts on either side of it.


Two things made it hard. The product problem: an app can get you into a room, but it can’t make you want to come back. The craft problem: Luma publishes no design system, so I had to rebuild one from screenshots — then add three features it has no pattern for, using a palette of seven greys and no accent colour.

Luma is excellent at the event itself. This redesign builds the parts on either side of it.


Two things made it hard. The product problem: an app can get you into a room, but it can’t make you want to come back. The craft problem: Luma publishes no design system, so I had to rebuild one from screenshots — then add three features it has no pattern for, using a palette of seven greys and no accent colour.

FINAL ARTIFACTS

Three flows, one per moment

Three flows, one per moment

Three flows, one per moment

Before · Onboarding — learning your own preferences

Questions most people are never asked: small room or large, seated or moving, alone or with someone.

Questions most people are never asked: small room or large, seated or moving, alone or with someone.

During · Event Buddy — SMS to in-app chat

A push becomes an invitation, a selection, a tone-calibrated message, an opened chat. Arriving alone stops meaning standing alone.

A push becomes an invitation, a selection, a tone-calibrated message, an opened chat. Arriving alone stops meaning standing alone.

After · Post-event reconnection

Only people who actually checked in are surfaced. Pick up to three. Neither is told unless both choose — no visible ask, so no rejection.

Only people who actually checked in are surfaced. Pick up to three. Neither is told unless both choose — no visible ask, so no rejection.

THE PROBLEM

The event is the easy part.

The event is the easy part.

The event is the easy part.

Before you go: what kind of room am I OK in?

While you’re there: I came alone — who do I talk to?

After: that was fun. But now there’s no reason to text anyone.

Before you go: what kind of room am I OK in?

While you’re there: I came alone — who do I talk to?

After: that was fun. But now there’s no reason to text anyone.

After the event

Luma has no screen here.

The product ends at checkout.

WHY IT MATTERS

30–50 hrs

of time together to make a casual friend. Over 200 for a close one. One event is never going to be enough.

of time together to make a casual friend. Over 200 for a close one. One event is never going to be enough.

Hall, 2018

1 in 3

adults in the U.S. feel lonely. One in four have nobody to lean on.

CDC

26%

of people know most of their neighbors. 12% know none of them.

Pew, 2025

Comparable

to smoking

Comparable to smoking

The Surgeon General says being dis-connected is as bad for you as smoking every day.

U.S. Surgeon General

Friendship takes repeat contact. One event can’t give you that. So the work is everything around the event.

Friendship takes repeat contact. One event can’t give you that. So the work is everything around the event.

WHERE THE MARKET STOPS

Stage

Who does it well

Who fails

Verdict

Discover

Meetup, Eventbrite, Posh

—

Solved

Match

Timeleft, 222, Pie

Bumble BFF (1:1 swiping)

Solved

Meet

Timeleft, 222, Pie

Eventbrite, Posh

Solved

Meet again

Heylo (existing groups only)

Timeleft, 222, Meetup

Partial

Maintain

Nobody

Everybody

Open

Every app is good at getting you into the room. None of them help before or after.

We ran our own survey before designing anything. It backed some of this up — and caught one thing I got wrong.

We ran our own survey before designing anything. It backed some of this up — and caught one thing I got wrong.

PRIMARY RESEARCH · EVENT PREFERENCES SURVEY · N=8

Eight people, eight different answers

Eight people, eight different answers

Eight people, eight different answers

We asked twelve questions and eight people answered. That’s a small group, so treat it as a hint rather than proof. Three answers were still very clear — and one of them showed I’d made a mistake.

We asked twelve questions and eight people answered. That’s a small group, so treat it as a hint rather than proof. Three answers were still very clear — and one of them showed I’d made a mistake.

4 + 4

Four preferred a buddy; four were comfortable attending either way.

Buddy pairing supports a preference, not a universal reluctance to attend alone.

8 of 8

picked food as something they want at an event.

Food/dining and weekend-evening timing both received 8/8 selections.

6 of 8

wanted an activity to do — more than open mingling (5) or high-energy (2).

People want something to do. Having a task makes it easier to talk.

7 of 8

picked an identity or access option: BIPOC friendly (3), 21+ (4), quiet / sensory-friendly (1). Only one person picked none.

picked an identity or access option: BIPOC friendly (3), 21+ (4), quiet / sensory-friendly (1). Only one person picked none.

No dedicated question in the app. Almost everyone asked for them.

Identity and access options were in the survey. In the prototype they never got a question — Q11 asks who you go with, and the completion portrait matches that.

Identity and access options were in the survey. In the prototype they never got a question — Q11 asks who you go with, and the completion portrait matches that.

No two respondents gave the same set of answers

All eight response profiles were different. This suggests preference variation within a small sample, not proof that every question needs its own onboarding screen.

All eight response profiles were different. This suggests preference variation within a small sample, not proof that every question needs its own onboarding screen.

Which questions earned their screen

These are the survey questions, not the app screens. Short bar = the question told us something. Long bar = everyone said the same thing.

These are the survey questions, not the app screens. Short bar = the question told us something. Long bar = everyone said the same thing.

Q10 · Frequency

Every couple of weeks

3/8

Q1 · Attendance style

Buddy / either — split 4–4

4/8

Q4 · Alcohol

Either is fine

4/8

Q6 · Seating

Seating available

4/8

Q11 · Accessibility & identity

21+ fine — but 7 of 8 chose something

4/8

Q12 · Cost

Willing to pay

4/8

Q2 · Event size

Small / intimate

5/8

Q5 · Vibe

Structured activities

6/8

Q3 · Setting

No preference

4/8

Q9 · Distance

Anywhere in the city

6/8

Q8 · Timing

Weekend evening — everyone

8/8

Q7 · Activities

Food / dining — everyone

8/8

Food/dining and weekend-evening timing both reached 8/8. Distance reached 6/8 and setting’s largest answer was 4/8. These distributions describe preferences, not usability. Survey Q11 addressed identity; prototype Q11 asks who you usually go with.

Food/dining and weekend-evening timing both reached 8/8. Distance reached 6/8 and setting’s largest answer was 4/8. These distributions describe preferences, not usability. Survey Q11 addressed identity; prototype Q11 asks who you usually go with.

Food was a shared starting point.

All eight selected food/dining and weekend-evening timing. Four preferred a buddy; four were comfortable either way. A shared meal is a promising starting point, not a tested outcome. My experience cooking professionally informed this direction; the prototype still needs evaluation.

All eight selected food/dining and weekend-evening timing. Four preferred a buddy; four were comfortable either way. A shared meal is a promising starting point, not a tested outcome. My experience cooking professionally informed this direction; the prototype still needs evaluation.

TEAM AND PROCESS

Three designers, one shared board, one disagreement worth keeping

Three designers, one shared board, one disagreement worth keeping

We worked out of a single FigJam board. Research was split three ways, then we synthesized together and I took the interaction design from there.

We worked out of a single FigJam board. Research was split three ways, then we synthesized together and I took the interaction design from there.

Navneet

Competitive analysis

Audited nine products — Timeleft, 222, Pie, Meetup, Heylo, Geneva, Bumble BFF, Eventbrite, Posh — with pros, cons and sources for each. This is where the “nobody solves Maintain” conclusion came from.

Audited nine products — Timeleft, 222, Pie, Meetup, Heylo, Geneva, Bumble BFF, Eventbrite, Posh — with pros, cons and sources for each. This is where the “nobody solves Maintain” conclusion came from.

Christina

Secondary research

Built the evidence base: Surgeon General, WHO, CDC, Pew. The friendship-hours figure that shapes the whole approach came out of her stack.

Built the evidence base: Surgeon General, WHO, CDC, Pew. The friendship-hours figure that shapes the whole approach came out of her stack.

Me

Synthesis and interaction design

Friendship-formation research, then the preference survey, IA, all three flows, the tag system and the visual system. I owned everything from the board to the screens.

Friendship-formation research, then the preference survey, IA, all three flows, the tag system and the visual system. I owned everything from the board to the screens.

How the work moved

01

Split research

Each of us took a lane so we weren’t reading the same papers. One board, three sections.

Each of us took a lane so we weren’t reading the same papers. One board, three sections.

02

Synthesize together

We mapped every product onto five stages — Discover, Match, Meet, Meet again, Maintain — and found the empty row.

We mapped every product onto five stages — Discover, Match, Meet, Meet again, Maintain — and found the empty row.

03

Test the assumption

I ran a twelve-question preference survey to check whether the tags we’d invented meant anything to real people.

I ran a twelve-question preference survey to check whether the tags we’d invented meant anything to real people.

04

Design, then audit

I built the flows, then went back to the board and checked the build against what we’d actually concluded. It didn’t match.

I built the flows, then went back to the board and checked the build against what we’d actually concluded. It didn’t match.

Where we disagreed, and who turned out to be right

The team proposed progressive onboarding: a few questions at signup and another at RSVP. I prototyped twelve upfront to collect preferences before the feed. Eight unique response profiles showed variation, but did not establish that twelve screens were usable or necessary. The length and question selection remain hypotheses to test against the team’s shorter, progressive approach.

The team proposed progressive onboarding: a few questions at signup and another at RSVP. I prototyped twelve upfront to collect preferences before the feed. Eight unique response profiles showed variation, but did not establish that twelve screens were usable or necessary. The length and question selection remain hypotheses to test against the team’s shorter, progressive approach.

INFORMATION ARCHITECTURE

Five tabs on paper. Four in the prototype.

Five tabs on paper. Four in the prototype.

Five tabs on paper. Four in the prototype.

Home

Upcoming events, continue a connection, familiar communities

Discover

Personalized discovery with vibe tags

Plans

Upcoming events, plans with buddies, past events

Connections (folded into Chat)

People met, small groups, event chats, familiar faces

Profile (not prototyped)

Interests, social preferences, privacy, notifications

The prototype uses Home · Discover · Plans · Chat. See the limitations block for the difference from the intended architecture.

The prototype uses Home · Discover · Plans · Chat. See the limitations block for the difference from the intended architecture.

INFORMATION ARCHITECTURE

Intended five tabs, mapped onto before / during / after

Home

Discover

Plans

Connections

Profile

BEFORE

Preference portrait. Continue a connection.

“Why it fits” on every card.

—

—

Social preferences. Identity and access tags.

DURING

Upcoming events. Familiar communities.

Personalized feed. Vibe tags on events.

Plans with a buddy.

Event Buddy. Familiar faces. Event chats.

—

AFTER

Continue a connection.

—

Past events. Reconnection.

People met. Small groups.

—

Plans and Chat carry the post-event flow in this prototype.

Plans and Chat carry the post-event flow in this prototype.

Features

Nine features, and what each one changes for the person walking in

Nine features, and what each one changes for the person walking in

Every feature answers a question the attendee is already asking themselves. The right column is the only test that matters.

Every feature answers a question the attendee is already asking themselves. The right column is the only test that matters.

BEFORE

Knowing what you want

Social preference onboarding

Twelve questions on room size, seating, vibe, alcohol, company.

You find out you are a small-room person. Nobody has asked you that before.

“Why it fits” card · not in prototype

Each suggested event was meant to state the two or three preferences it matches.

Prototyped instead: vibe tags on the card, rather than a personalized explanation.

Prototyped instead: vibe tags on the card, rather than a personalized explanation.

Preference profile · not in prototype

Answers were meant to live in a Profile tab, editable any time.

They appear once, on the completion portrait. There is no Profile tab to go back to.

DURING

Meeting someone so you are not there alone

Event Buddy

A push invites you to pair with another attendee before the event.

An option to arrive with someone: four respondents preferred a buddy, while four were comfortable either way.

Familiar faces · not in prototype

The event page was meant to show who you have already met.

No event-detail screen in the prototype does this. You meet people through Event Buddy and Plans instead.

No event-detail screen in the prototype does this. You meet people through Event Buddy and Plans instead.

Identity & access tags · no dedicated screen

BIPOC friendly, LGBTQIA+ friendly, quiet / sensory-friendly, wheelchair accessible.

Seven of eight asked for this. There is no dedicated question — Q11 in the flow asks who you go with.

Seven of eight asked for this. There is no dedicated question — Q11 in the flow asks who you go with.

AFTER

Staying in touch

Mutual reconnection

2–12 hours after, pick up to three people who checked in.

You can reach out with no risk. Neither of you is told unless you both choose.

Connections · not a tab

People met live in Plans → Recently Met, not a Connections tab.

The person you liked is still findable next week, not buried in a thread.

Plan Together

Turn a connection into a next event in two taps.

A reason to see someone again, which is the thing that was always missing.

DESIGN SYSTEM

The tag vocabulary is the product

The tag vocabulary is the product

Twenty-four tags in five groups, drawn at five sizes. These are the same words a user picks during onboarding — which is how a preference becomes a match instead of a stored answer.

Twenty-four tags in five groups, drawn at five sizes. These are the same words a user picks during onboarding — which is how a preference becomes a match instead of a stored answer.

Location

“City Name”

District

Theme

Tech

Food & Drink

AI

Arts & Craft

Climate

Fitness

Wellness

Crypto

Event Setting

Small / Intimate

Large / Lively

Indoor

Outdoor

Atmosphere

Alcohol-free

High-energy

Low-key

Mingling

Structured

Seating

Time

Weekend

Weekday

Daytime

Evenings

Onboarding asks about size, seating, atmosphere and timing. Event chips use those words. Identity tags were supposed to be in that same vocabulary — they never got a question of their own.

Onboarding asks about size, seating, atmosphere and timing. Event chips use those words. Identity tags were supposed to be in that same vocabulary — they never got a question of their own.

Chip scale

46 / 39 / 29 / 22 px

Five sizes of the same tag row, from event hero down to dense list.

Avatar scale

72 / 47 / 38 / 32 px

Five illustrated faces at four sizes — used for familiar faces and buddy pairing.

Thumbnails

60 / 40 px

Event imagery at two densities.

Glass Nav

267 × 52

Home · Discover · Plans · Chat, floating over content.

24

tags in five groups, matching the onboarding question set

15

screens rebuilt on one frame structure — status bar, header, body, absolute CTA

171

hardcoded fills and strokes rebound to variables, no colour values changed

25

colour tokens; nine run the entire onboarding flow

The colour tokens are a reconstruction, not Luma’s published palette — the real product has no fixed accent set. A host picks one theme colour per event and the system derives the rest, adjusted for contrast.

The colour tokens are a reconstruction, not Luma’s published palette — the real product has no fixed accent set. A host picks one theme colour per event and the system derives the rest, adjusted for contrast.

HOW IT GOT MADE

Three passes on the onboarding flow

Three passes on the onboarding flow

Three iterations developed the onboarding concept. The preference survey informed the content; usability and completion length remain untested.

Three iterations developed the onboarding concept. The preference survey informed the content; usability and completion length remain untested.

Pass 01 · Sketch

What questions exist and in what order. Paper, no interface.

The order kept changing because I had no reason to prefer one.

Pass 02 · Wireframe

One question per screen, progress bar, back arrow. Structure only.

Twelve screens felt long. I could not tell if that was a problem or a feature.

Pass 03 · Built

Real type, real tokens, selection by tint and border.

The survey showed preference variation; it did not test onboarding length.

Preference variation is evidence for personalization, not validation of twelve screens.

KEY DESIGN MOMENTS

Five decisions, and the versions I rejected

Five decisions, and the versions I rejected

Five decisions, and the versions I rejected

01

Twelve questions is a lot. Is the length the problem?

What happened

Team research said five questions. I built twelve.

Why it matters

Eight unique response profiles suggest varied preferences. They do not validate the usability or length of a twelve-screen onboarding flow; that remains an untested hypothesis.

Eight unique response profiles suggest varied preferences. They do not validate the usability or length of a twelve-screen onboarding flow; that remains an untested hypothesis.

The decision

Prototype twelve questions and a completion portrait, then compare against progressive onboarding in usability testing.

Prototype twelve questions and a completion portrait, then compare against progressive onboarding in usability testing.

02

Which questions actually describe a person?

What happened

Four survey questions were logistics. Identity never got its own screen — Q11 in the prototype is “who do you usually go with?”

Four survey questions were logistics. Identity never got its own screen — Q11 in the prototype is “who do you usually go with?”

Why it matters

Distance: six identical answers. Setting: four “no preference.” Identity tags: seven of eight selected one. I cut the wrong four.

Distance: six identical answers. Setting: four “no preference.” Identity tags: seven of eight selected one. I cut the wrong four.

The decision

Cut distance and setting. Give identity its own question — it never made it into the twelve.

Cut distance and setting. Give identity its own question — it never made it into the twelve.

03

How should selection look in a product with no color?

What happened

Luma ships 40+ themes — then strips all of them on iOS for consistency.

Why it matters

So a neutral shell is the correct reading, not a shortcut. My palette is seven greys; selection cannot be a hue.

So a neutral shell is the correct reading, not a shortcut. My palette is seven greys; selection cannot be a hue.

The decision

Selected = 3% tint + black border. No accent color in onboarding.

04

Chat, or Connections?

Luma today

Three tabs: Home, Discover, Chat. No Plans, no Connections, no Profile.

What happened

Five tabs became four. Connections merged into Chat, Profile dropped entirely. The nav is Home · Discover · Plans · Chat.

Five tabs became four. Connections merged into Chat, Profile dropped entirely. The nav is Home · Discover · Plans · Chat.

Why it matters

Chat is threads that already exist. Connections is people you met. Profile is the answers. I deleted two thirds of the thesis from the nav.

Chat is threads that already exist. Connections is people you met. Profile is the answers. I deleted two thirds of the thesis from the nav.

05

Adding features to a system that was never published

Luma’s own option row

Identity tags — borrowed from category chips

Event Buddy — invented

Connections — invented

What happened

Luma publishes no design system and no component library. I rebuilt one from screenshots — then had to add three features Luma has no pattern for at all.

Luma publishes no design system and no component library. I rebuilt one from screenshots — then had to add three features Luma has no pattern for at all.

Why it matters

“Match Luma” has no single answer: the product is a theming engine with 40+ themes, and it strips all of them on iOS. So the target was the neutral iOS shell — seven greys, no accent, selection by tint and border. That leaves almost nothing to differentiate a new feature with. Invent too much and it stops being Luma; invent too little and Event Buddy reads as another settings row.

“Match Luma” has no single answer: the product is a theming engine with 40+ themes, and it strips all of them on iOS. So the target was the neutral iOS shell — seven greys, no accent, selection by tint and border. That leaves almost nothing to differentiate a new feature with. Invent too much and it stops being Luma; invent too little and Event Buddy reads as another settings row.

The decision

New features borrow existing primitives and earn their novelty from the flow, not the visual language. Identity tags reuse the category-chip pattern. Event Buddy reuses the messages surface. Nothing new got a new color.

New features borrow existing primitives and earn their novelty from the flow, not the visual language. Identity tags reuse the category-chip pattern. Event Buddy reuses the messages surface. Nothing new got a new color.

IMPACT AND REFLECTIONS

The survey told me which questions didn’t earn a screen.

The survey told me which questions didn’t earn a screen.

The survey told me which questions didn’t earn a screen.

These are prototype deliverables, not shipped outcomes. We ran four interviews and a survey of eight. Four preferred a buddy and four were comfortable either way; food/dining and weekend-evening timing both received 8/8 selections. Preference variation informed the design, but onboarding length and usability remain untested. No shipped metrics are claimed.

These are prototype deliverables, not shipped outcomes. We ran four interviews and a survey of eight. Four preferred a buddy and four were comfortable either way; food/dining and weekend-evening timing both received 8/8 selections. Preference variation informed the design, but onboarding length and usability remain untested. No shipped metrics are claimed.

Prototype deliverables

3 flows, 15 screens, 24 tags, 25 tokens

Limitations · not prototyped

“Why it fits”, Profile tab, Connections tab, identity question

Would cut

Reassess repetitive questions through usability testing

Would restore

A real identity question — 7 of 8 asked, Q11 is still “who do you go with?”

What I would measure

Do people revisit their preferences? Do solo attendees arrive with a buddy? Return-attendance within 30 days. Registration counts say nothing.

Do people revisit their preferences? Do solo attendees arrive with a buddy? Return-attendance within 30 days. Registration counts say nothing.

What I got wrong

I designed the input before the payoff, and stopped consulting the research once designing started. Research is something you return to during the work, not a phase you finish before it.

I designed the input before the payoff, and stopped consulting the research once designing started. Research is something you return to during the work, not a phase you finish before it.

More Projects

Created by

Mj

in

Framer

Resume

Created by

Mj

in

Framer

Resume

MJ

Luma

Problem

Overview

Research

Solution

Impact

Reflection

Ui/Ux Redesign

2026 · Case 01/04