What i did
UX Strategy
Interaction Design
UI & Visual Design
Product Discovery

AI personalized recipes for organic food box subscribers

I led the design of AI Powered Recipes, a feature for Diferente’s organic food box platform that uses generative AI to create personalized recipes based on each user’s box contents. The work combined AI prompt engineering, UX strategy, and visual design to deliver a dynamic recipe experience that increased in-app engagement and product value.
What i did
UX Strategy
Interaction Design
UI & Visual Design
Product Discovery
The challenge
Help people prepare better meals with a specific selection of fruits and vegetables
Diferente's organic food boxes contain a wide variety of vegetables, fruits, and seasonings, and people didn't know what to prepare with them.

One of the most common topics in interviews and comments was regarding food preparation. To solve the issue of food preparation, we knew we needed a feature that would be personalized for each person's box.

Since some people have dietary restrictions, each box ends up being unique, so we couldn't create a generic solution that would work to everyone.
01.
People didn’t know what meal to prepare with the variety of fruits and vegetables they received.
02.
Each person's food box was unique so we couldn’t create generic recipes to solve all the problems.
Our approach
Creating tailored recipes for each person using Generative AI
Using ChatGPT to generate personalized recipes, we developed a feature in the app that tailors suggestions 100% to each user's organic box.

We introduced a section in the mobile app called "What do you want to prepare today?" featuring at least 20 AI-generated recipes. This section refreshes with new recipes every time users receive a new food box.

Users can choose between recipes for specific items in their box or recipes for the entire box. Additionally, we included a section dedicated to other products we sell, beyond fruits and vegetables.
Boosting the sales of additional products in the recipe page
One of the goals of this initiative was to boost sales of our other products, such as eggs, mushrooms, and plant-based milk.

To support this, we added a component that displayed the ingredients used in each recipe along with their prices. With just a few clicks, users could add these products to their box.
AI prompts as handoff
To ensure that ChatGPT generated recipe data accurately, it was also part of my job to test and validate various types of prompts.

We aimed to find the ideal prompt that developers could use to automate the functionality.
We learned the hard way that some fruits and vegetables share the same names as cities in Brazil
The results and key learnings
21% feature adoption in week one
Within the first week of launch, 21% of the customer base used the AI recipe feature, incorporating personalized meal recommendations into their weekly planning.
+11% additional products purchases
Personalized recipe suggestions encouraged users to complement their meal plans, resulting in an 11% increase in add-on product purchases.
Prompt iteration was essential
The initial prompts used with ChatGPT didn’t always produce the expected results and occasionally led to confusing outputs. Through quick iterations during the first weeks, we refined the prompts to improve clarity and consistency.