Deciding what to cook for dinner sounds trivial, until it becomes a daily source of decision fatigue. Our client wanted to solve exactly that. Their goal was to build a mobile application for the Swedish market that takes the friction out of meal planning: something genuinely personalized, easy to use, and smart enough to get better over time. The app lets users build weekly meal plans, explore recipes matched to their tastes, and automatically generate shopping lists based on whatever meals they've picked.
What sets it apart is that it doesn't rely on static, one-size-fits-all suggestions. The platform continuously adapts based on user behavior, dietary preferences, and interaction history to keep recommendations relevant and useful. Our team handled end-to-end product development: the mobile app, a recommendation engine, and a CMS for recipe management.
The client's ambition was to create a meal-planning experience that actually understood the person using it. That meant tackling a recommendation problem with real complexity. The system needed to account for individual preferences, dietary restrictions, and eating habits, while keeping suggestions varied enough to stay interesting over time. Getting the balance right mattered. A recommendation engine that's too narrow becomes repetitive; one that's too broad stops feeling personal.
Some failure modes were straightforward to define, like suggesting meat dishes to vegans or flagging seafood recipes for users with allergies. Preventing these cases reliably at scale was a different story. The engine had to handle these constraints consistently, without letting them collapse the range of what it could suggest. Usability raised its own set of challenges. The app was aimed at everyday consumers, not power users willing to configure complex filters or scroll through endless recipe lists. Everything from discovering meals and building a weekly plan to generating a shopping list had to feel immediate and low effort, regardless of how much was happening under the hood.
On the technical side, the recommendation engine had to get smarter over time on its own, learning from user interactions without putting that burden on the user.
We built a cross-platform mobile application centered on personalized meal planning and intelligent recipe discovery with a recommendation engine that continuously adapts to individual preferences and behavior.
At the core of the platform is a vector-based recipe search system powered by Azure AI Search. During onboarding, users build a preference profile that gets translated into a vector representation. Recipes in the system are processed the same way, enabling the platform to surface the most relevant matches through vector similarity search.
From there, the recommendation logic evolves with use. Every interaction within the app contributes to refining the user profile, gradually improving recommendation accuracy while maintaining enough variety to keep suggestions from becoming repetitive.
The project scope covered:
To support richer content experiences, the platform integrated OpenAI APIs for text generation and NanoBanana for image generation workflows. Application infrastructure was hosted and managed via Azure App Service.
The client received a fully functional mobile application that meaningfully simplified meal planning, cooking, and shopping.
The platform delivers personalized recipe recommendations based on user preferences, dietary restrictions, and behavioral patterns, helping users find relevant meals quickly without manually sifting through large recipe collections.
Beyond the app itself, the project laid a scalable foundation for future growth: a custom recommendation engine, AI-powered content workflows, and a dedicated CMS for recipe management.
Since launch, the product has gone through multiple rounds of iteration, with ongoing refinements focused on recommendation quality, usability, and overall user experience.
