Mateus Peruzzo

Brinda

Research · UX/UI · 2026


Overview

Brinda answers one question: what should I drink with this?

You type in what you’re about to eat, and you get three suggestions — the classic one, an original one, and one drawn from wherever the dish comes from. Each comes with a short note explaining why it works, because knowing why is what lets you make the next choice on your own.

The app is built for people who cook out of curiosity. Someone making feijoada on a Sunday who wonders whether there’s something better than the beer already in the fridge. Someone who has read that white goes with fish and red with meat, and suspects that’s not the whole story. You don’t need to know anything about wine to use it, and you won’t be talked down to for not knowing.

Pairing has always been guarded knowledge — passed between sommeliers, printed in books that assume you already speak the language. Brinda is an attempt to hand it over in plain words, one dish at a time.

Where it goes next

The same idea works inside a shop or a restaurant. A supermarket has a wine aisle and no one standing in it to help; a restaurant has a list and a server who may or may not know it well. Brinda could work from an actual inventory — recommending from what’s on the shelf that night, not from an ideal cellar that nobody has.

That’s the longer arc: the app in your hand first, and eventually the same knowledge available at the exact moment someone is standing in front of a shelf, deciding.


User Research

Before designing anything, I wanted to know how people actually decide what to drink with a meal — not what they think they should do, but what happens at the table. Six conversations later, most of what I had assumed turned out to be wrong.


Method

Six conversations, over WhatsApp and in person, before any screen existed. None of the participants work in food or beverage. They were in Brazil and Europe, across a range of ages, and included one person who no longer drinks alcohol.

Some are people I know. That made the conversations franker, and it narrows how far the findings generalise — worth stating either way.

I chose conversations over a survey because I wanted to hear people describe their own reasoning, not pick from options I had written for them. Quotes below are translated from Portuguese except where the participant spoke English.


What they said

They care, and they’ve given up anyway

"It matters a lot, but we set it aside because we don't know how."

"I believe it's very important, even if it's not within my reality."

Participants rated pairing as important and then described doing it only on special occasions. The gap between what they valued and what they did was the widest finding in the research.

They want the mechanism, not the answer

"What's the fundamental difference between a Merlot and a Cabernet Sauvignon? And how does that difference apply to pairing with a meal?"

"Price and knowledge — so I don't have to depend on the opinion of the staff."

All six asked to understand why a pairing works. Not to learn for its own sake: they wanted to be able to choose on their own, in front of a wine list or in a supermarket aisle.

The repertoire stops at two rules

"Knowing what works beyond fish and red meat."

Every participant described the same pair, unprompted: white with fish, red with meat. Several named it themselves as the limit of what they knew.

Non-alcoholic is underserved, not absent

"There aren't enough non-alcoholic options that aren't too sweet."

Five of six wanted non-alcoholic pairings. The complaint was not that they don’t exist — it was that the ones that do are either sugary or dull.


Pain points

  1. 1 — Interest without a route. People value pairing and give up on it anyway, because they don’t know where to start.
  2. 2 — No explanation, no autonomy. Every participant asked to understand why a pairing works — so they could choose without help next time.
  3. 3 — Suggestions they can’t act on. Price and availability decide the final choice. A technically perfect recommendation is useless if the bottle isn’t on the shelf.
  4. 4 — A repertoire stuck at two rules. White with fish, red with meat. Participants named this themselves and asked for a way past it.

What it changed

Where the three categories came from

I asked each participant which they would want: a classic pairing, an original one, or one drawn from the dish’s region. The answers didn’t converge on a category. They converged on context.

"If I'm travelling and getting to know a place, regional. In a well-reviewed restaurant, maybe the original one. At home, classic."

That is why Brinda returns all three at once instead of asking users to pick a mode. The same person wants a different answer depending on where they are — so the app stops asking and simply answers three ways.

The second pain point shaped the card itself. Because every participant asked why, the explanation is not a detail behind a tap: it sits on the card, in the second colour tier, above the origin.

Read the full interviews


Wireframes

The first structure, built on what the interviews had shown. Each pair shows the original and what replaced it after testing.

Home: o wireframe cinza à esquerda e, à direita, a tela final do Brinda com a busca e o Journal. Resultados: o wireframe com três blocos iguais e, à direita, os três cards — Clássica, Regional e Original — já com o motivo de cada um. Detalhe da harmonização: o wireframe com “add to my list” e, à direita, a tela final com Resumo, Detalhes e Ciência.

Product architecture

Brinda is built around a single action, and everything else supports it. You type a dish, you get three drinks, and the reason each one works.

Four tabs

Home is the search and the result. Almost all the time in the app is spent here.

Saved holds the pairings you wrote yourself, after the meal — the dish, the drink, and your own note on why it worked. No button on a result files one here, and that is deliberate: bookmarking a suggestion keeps you leaning on the app, while writing down what actually happened teaches you to choose. The interviews asked for the second one.

Journal is a feature of one pairing a week, photographed and written by hand, chosen from submissions. It exists because the research showed people wanted to learn, not just be answered.

Profile is the lightest of the four today: the account and its settings. It’s also where the app is headed — a place to register what’s actually in your cellar, so the suggestions come from bottles you already own.

Three answers instead of one question

The obvious design would ask what kind of pairing you want, then give you one. Brinda gives all three at once and doesn’t ask.

That came straight from the interviews. Nobody picked a favourite category — they picked differently depending on where they were. Travelling, they wanted the regional one. In a good restaurant, the original one. At home, the classic. Asking would mean making people predict their own mood; answering three ways lets them recognise it.

Preferences belong to the meal, not to the person

There’s no saved preference profile. Filters — wine, beer, cocktails, tea, juice, no alcohol — live on each search and reset afterwards.

That’s deliberate. You might be cooking for a friend who doesn’t drink, or for a table where one person does and another doesn’t. A profile would assume every meal is yours. Treating the filter as part of the question instead of part of the person keeps the app useful when you’re feeding someone else.

Where the inventory idea leads

A cellar, a supermarket aisle and a restaurant list are the same problem: a real, limited stock, and a person standing in front of it without help. The app starts by suggesting from everything that exists. The version worth building suggests from what’s actually there.

Dish Classic Regional Original Saved optional drink filter You write the note yourself One dish, three pairings, and the reason behind each

Colour

The ground is a warm off-white rather than pure white. Food photography sits badly on neutral white — fat, crust and fruit skin all read colder than they are.

Hierarchy is carried by value, not hue. The brand green carries the answer and the reason it works together, at the strongest value on the page: explaining why a pairing works is the product, not a footnote to it. Below them one warm green-grey runs at three steps — graphite for serving detail, mid for metadata and inactive states, light for borders. Ordered by how dark they are, the hierarchy holds in greyscale and never competes with the photographs beside it.

The brand green carries the ink. A near-black green reads as black at text sizes, and that is the argument for it rather than against it — it warms every screen at no cost, because the eye takes it as a dark neutral. Pure black left the palette with it: on cream it reads as a hole.

Five mid-greys were doing one job, within four points of each other and indistinguishable on screen. The same sentence carried two of them in two different components. They are now one token. Collapsing them went one step too far, though: the explanation of why a pairing works came out the same weight as the region it sat next to. It did not need a new grey, it needed promoting.

The peach is one hue at two strengths — a tint for headers, full strength for notes that interrupt.

One failure found late. The border tone had been corrected to clear 3:1, the bar for interface elements, and was then reused for 13px tab labels, where the bar is 4.5:1. It measured 3.2:1. A colour is not accessible on its own, only in a role: it is now interface-only, and inactive labels take the body grey at 5.4:1.

The active navigation state was a pale tint at 1.3:1. Darkening it enough to register pushed the label above it below the minimum — one problem displacing the other. So the state is marked by icon shape, filled when active, which survives greyscale and colour blindness.


Colour palette

  • #FFFBEFPage ground
  • #FFFFFFCards
  • #FEF5E5Headers
  • #FCE9D3Notes
  • #EFEBE0Loading
  • #083306The answer and the why
  • #45493DServing, buttons
  • #636A57Metadata, inactive
  • #8B9174Borders only

Type System

Two screens, twelve type sizes, and no rule for when to use the serif and when to use the sans. People paused on the home screen before doing anything — nothing told them where to look first.


What was there

74 · 22 · 20 · 18 · 18 · 16 · 15 · 14 · 13 · 13 · 12 · 11

Challenge

Twelve sizes across two screens, some only a pixel apart. The three pairing categories were set three different ways.

Solution

Eight sizes, fourteen roles, every one with a fixed line height. The three categories now share a single role.

The main button was sans, the “see more” link beside it was serif, and the navigation labels were a third thing, small caps at 11px. Three different answers to the same question, on one screen.

Changing the typeface

Challenge

Cormorant Garamond looked right for a wine app, but its thin strokes vanish at small sizes — and small is where this app lives.

Solution

Merriweather and Libre Franklin, both drawn for screens. Less character, more read.

It is a real trade-off, and Cormorant had more character. But the research was about people not understanding what they were looking at, so I would rather have the font that is easier to read.

The rule

Challenge

No reason behind any size. Every new screen quietly added another number to the pile.

Solution

Base 16, ratio 1.25 going up. Below the base, 13, 12 and 11 are set by hand — and the page says so instead of pretending.

The wordmark on the home screen turned out to be the seventh step of the ratio and the one in the header the fourth — I only found that out by measuring, which is a good sign that the ratio was doing real work rather than decorating a slide.

Eight sizes, fourteen roles. A role is a family, a weight, a size and a line height together, so one size carries more than one job: 13 is the tasting note in the serif, the tab label in the sans medium, and the sheet paragraph in the sans regular. Counting sizes alone tells you very little about whether a type system is under control.

Hierarchy

Challenge

Nothing told people where to look first. Weight, colour and spacing were doing none of the work.

Solution

Four steps of one green-grey, serif for content, sans for anything you tap. Nothing depends on colour alone.

Weight. The serif goes bold only for the wordmark — Merriweather was drawn to take a bold without getting muddy, so one step of weight can mark a level that size alone would not. The sans does the rest: medium for screen titles and tabs, bold for the single action on a screen.

Colour. Four steps of one warm green-grey, named by importance rather than by hex value: the brand green carries both the answer and the reason it works, graphite the serving detail, mid grey the metadata, and the lightest step is reserved for borders and never used for text.

Spacing. Inside a card, the drink name and its note sit 7px apart. Between cards it is 20px, so you read the card as one block before you read the words in it.

Family. The serif holds content. The sans holds anything you tap, or anything that is a label.

The reasoning sits at the same level as the answer, because every person I interviewed asked why a pairing works. That order follows the research, not the convention.

RoleSpecSample
WordmarkMerriweather Bold 39/44 Brinda
Screen titleLibre Franklin Medium 25/38 Minha adega
SectionMerriweather Regular 20/28 Feijoada
FeatureMerriweather Medium 16/28 Douro Reserva
BodyMerriweather Regular 16/24 Diga o que vai comer
ActionLibre Franklin Bold 16/20 Ver o que combina
NoteMerriweather Regular 13/22 O tanino firme corta a gordura
Sheet bodyLibre Franklin Regular 13/22 Valem para esta sessão
TabLibre Franklin Medium 13/18 Resumo · Detalhes · Ciência
SupportLibre Franklin Regular 13/18 Ver mais
Section labelLibre Franklin Light 12/30 ONDE BUSCAR
NavigationLibre Franklin Regular 12/16 Harmonizar
Chip labelLibre Franklin Medium 12/16 DESTILADOS
MetadataLibre Franklin Regular 11/16 Douro, Portugal

Applied

The home screen before and after. Same layout, same content — only the type system changed.

A home antes e depois: à esquerda o Pairing Journal em versalete e a busca em Cormorant; à direita, Journal em Merriweather e os mesmos elementos com a escala nova. Os resultados antes e depois: à esquerda os três cards sem hierarquia clara; à direita, nome, motivo e origem em três níveis distintos.

Iconography

Every icon is drawn on the same 24×24 grid with a 2px stroke. The active tab uses the filled version, the others the outline — the same silhouette either way, so nothing grows or shifts between tabs. Shape carries the state because colour couldn’t: the pale peach background measured 1.3:1 against the bar, and darkening it enough to register pushed the label below the contrast minimum. Fixing one broke the other. Filled against outlined reads at any contrast, in greyscale, and without colour vision.

As duas barras de navegação lado a lado: à esquerda os ícones em contorno, à direita os mesmos preenchidos. Os quatro ícones em contorno, ampliados: casa, marcador, jornal e pessoa. Os mesmos quatro ícones preenchidos, com a silhueta idêntica à da versão em contorno.

Sticker sheet

Fifty-one pieces, built from the tokens on the Colour page and the scale on the Type page. Every caption carries the measured value, not the intention.

Brand and chrome: o logo Brinda, a status bar e os cabeçalhos de Home, seção, Clássica e Minha Adega, com a tab bar nos estados inativo e ativo. Buttons: o botão primário nos estados normal, pressionado e desabilitado, mais os botões de adicionar, fechar, sair da conta e mais opções. Input and selection: campo de busca vazio e preenchido, campo de formulário, slot de foto, checkbox, switch, seletor de álcool e o ícone de filtro limpo e ativo. Chips, labels and tabs: os chips da adega ativo e inativo, o rótulo de categoria, os selos do Journal e da adega, e as abas de salvos e de explicação. Cards: o bloco de resultados normal e pressionado, cards de busca salva, de journal, de autor, de ajustes, da adega, de envio, o card de adicionar tracejado, o cabeçalho de resultados e a caixa de explicação. Loading and error: o card de carregamento, os três pontos âmbar e verde da busca, os pontos inativos do erro e o chip com erro.

Two rules

Pressed always darkens, never lightens. Every pressed state is the default tone one step down in luminance. Lightening reads as fading; darkening reads as pushed in.

Disabled exists only where the action can be unavailable. The primary button has it, because the search field can be empty. Links, chips and tabs do not — they never appear when they cannot be used, and a state that never occurs is one more thing to maintain for nothing.


In use

Uma tábua de madeira com tomates, manjericão, alho, farinha e um ovo quebrado, e o Brinda aberto na tela inicial no meio da bancada.
Cooking at home, and feeling confident about what to offer the friends coming over.
Uma mão segurando o celular com as três harmonizações para feijoada, diante das prateleiras de vinho de uma loja.
Invited to dinner at someone else’s table, and knowing exactly what to bring along.
Um prato de feijoada servido numa mesa posta, com uma taça de Amontillado ao lado e o celular mostrando a explicação da harmonização.
Sitting in a restaurant, and understanding why the pairing in front of you works.
O fecho da marca: Brinda em verde sobre creme, o campo de busca com “what are you cooking today?” e a assinatura “Beyond the obvious”.
Back to top ↑