Making “what should I cook?” easier to answer.
Finding a recipe isn't the hard part. Choosing one is.
Recipe apps are great at giving people options. But when you’re deciding what to cook after a long day, more options can mean more work. You might have ingredients at home, half an hour to spare, a dietary preference, or a craving you can’t quite turn into a search query. And once you’ve found something, there’s still the work of saving it, organising it and figuring out when to cook it.
I wanted Yeum to reduce the decisions around cooking rather than simply adding more recipes.
Three decisions, one cooking journey.
What can I make? Start with what is already in the kitchen.
What do I feel like eating? Describe it naturally instead of building a search query.
What should I cook this week? Turn individual choices into a plan.
Gigi, making the planner easier to fill.
The planner worked, but filling it still felt like a chore: search → filter → save → assign → repeat. The things people plan around, like time, diet, ingredients and cravings, are easy to say out loud but tedious to express through filters. So I turned that filling-in step into a conversation.
For example: “High-protein vegetarian, mostly quick dinners under 30 minutes. I love Mediterranean flavours.” Instead of building the perfect search query, users simply say what they mean.
Three rules for Gigi.
01. Reuse familiar patterns. Gigi answers with Yeum’s existing recipe-card anatomy instead of a wall of generated text.
02. Keep the scope narrow. Gigi is a meal-planning assistant, not a general-purpose chatbot. Its role is clear.
03. Keep users in control. Gigi suggests. The planner remains editable. Users can change, remove or ignore anything it creates.
AI is a faster way into the product, not a replacement for it.
A recipe should help you cook, not just help you read.
Once someone chooses a recipe, they’re no longer browsing. They’re cooking. So the recipe experience brings the practical information forward: time, nutrition, ingredients, instructions.
From individual recipes to a week that makes sense.
A recipe solves one meal. The planner helps organise everything around it. I designed it so users can move between individual meals and the bigger picture of their week, while keeping changes editable in place.
Cooking isn't only about planning.
Yeum also has a social discovery layer for people who find inspiration through creators and shared recipes. Rather than making that content the centre of the product, I treated it as another route into discovery, something users can browse, save and return to when they need a spark.
When users know what they want, search stays simple.
AI isn’t meant to replace conventional discovery. Familiar categories like breakfast, lunch, dinner, desserts and healthy meals are still there for when someone already has something specific in mind.
Search when I know what I want. Gigi when I don’t know how to ask for it.
A small system built for consistency.
As Yeum expanded across discovery, recipes, planning and Gigi, I needed a shared visual language that could work across all of them. I created a lightweight system around reusable foundations and components so new screens could be designed faster without every interaction becoming a one-off.
Visual foundations: a restrained palette and clear type hierarchy keep the experience warm, readable and consistent. Forest green establishes hierarchy, cream keeps the interface warm, burnt orange draws attention to actions.
One product, three ways to decide.
I know what I have: ingredient-led discovery.
I know what I’m looking for: search.
I know what I want but can’t easily express it: Gigi.
Three discovery modes, and nobody is forced into the AI one.
Discover → Decide → Plan → Cook.
What I learned
Yeum started as a recipe discovery problem, but the more interesting problem was the decision that happens before someone searches. A product can have great search, powerful filters and thousands of recipes and still leave someone thinking: “Okay, but what should I actually make?”
Gigi taught me that AI doesn’t always need to replace an existing interaction. Sometimes its best role is to remove the work between intention and action. The goal wasn’t to help people find more recipes. It was to help people decide what to cook.
