Mobile App for creating dishes with the help of AI
Challenge
SmartChef AI generates cooking recipes with artificial intelligence, for people who know their way around a kitchen as well as for those who are starting out.
The brief was unusual in one respect: the app was to run entirely on AI. Not only the text of the recipe, but the photo of the finished dish as well, had to be generated rather than taken from a database or a stock library. That means the whole product stands or falls on the quality of the generation and on how quickly it comes back, because someone deciding what to cook will not wait.
Alongside the app, the client needed an administrative panel.
SmartChef AI generates cooking recipes with artificial intelligence, for people who know their way around a kitchen as well as for those who are starting out.
The brief was unusual in one respect: the app was to run entirely on AI. Not only the text of the recipe, but the photo of the finished dish as well, had to be generated rather than taken from a database or a stock library. That means the whole product stands or falls on the quality of the generation and on how quickly it comes back, because someone deciding what to cook will not wait.
Alongside the app, the client needed an administrative panel.
Solution
Mobitouch designed the app and built it in Flutter, with Firebase on the server side, Azure OpenAI Service and GPT-4 for the recipes, DALL·E for the images, and an admin panel in React.js and Node.js.
A recipe starts from what the user has. She selects a few ingredients and sets the conditions that matter to her: difficulty, cooking time, whether ingredients outside the chosen list are allowed, and dietary preferences such as vegan, keto or paleo. The app returns several recipes within moments, each with an illustrative photo of the dish generated for it.
Those options are what turn a generic generator into something usable. The same set of ingredients produces a different answer for someone with thirty minutes than for someone cooking for the evening, and the dietary settings keep the results inside what the user actually eats.
Recipes go to favourites, so a dish that worked does not have to be found again, and over time each user builds a personal cookbook.
The admin panel covers users, categories and ingredients, and its AI module lets the administrator follow user activity and the generated recipes for analysis.
Mobitouch designed the app and built it in Flutter, with Firebase on the server side, Azure OpenAI Service and GPT-4 for the recipes, DALL·E for the images, and an admin panel in React.js and Node.js.
A recipe starts from what the user has. She selects a few ingredients and sets the conditions that matter to her: difficulty, cooking time, whether ingredients outside the chosen list are allowed, and dietary preferences such as vegan, keto or paleo. The app returns several recipes within moments, each with an illustrative photo of the dish generated for it.
Those options are what turn a generic generator into something usable. The same set of ingredients produces a different answer for someone with thirty minutes than for someone cooking for the evening, and the dietary settings keep the results inside what the user actually eats.
Recipes go to favourites, so a dish that worked does not have to be found again, and over time each user builds a personal cookbook.
The admin panel covers users, categories and ingredients, and its AI module lets the administrator follow user activity and the generated recipes for analysis.
Results
SmartChef AI answers the question of what to cook from whatever is in the kitchen. A few ingredients and a couple of settings produce several complete recipes in seconds, each with its own generated image of the dish.
The favourites and the personal cookbook turn single generations into something the user returns to, rather than a novelty tried once.
On the client's side, the admin panel keeps users, categories and ingredients under control and gives insight into how the app is used and what it generates.
SmartChef AI answers the question of what to cook from whatever is in the kitchen. A few ingredients and a couple of settings produce several complete recipes in seconds, each with its own generated image of the dish.
The favourites and the personal cookbook turn single generations into something the user returns to, rather than a novelty tried once.
On the client's side, the admin panel keeps users, categories and ingredients under control and gives insight into how the app is used and what it generates.