Literally Cooking

Effortlessly plan and cook for large gatherings with our recipe app that combines multiple dishes into a streamlined workflow, saving time and ensuring every meal is ready to serve together.

Comment 1

Inspiration

Have you ever struggled to cook multiple dishes at once for a big gathering, unsure of the order to prepare everything or how to time it so all the food is hot and ready together? Introducing Literally Cooking, the ultimate tool to streamline your cooking process—perfect for the holiday season. Say goodbye to cold vegetables while waiting for the main dish; with Literally Cooking, every plate comes together on time.

What it does

Literally Cooking lets users say which two dishes they want to make. Our AI model processes that information to determine the most efficient way to cook two meals simultaneously. It then generates a "mega recipe" that is output to the user. This product allows for energy efficiency by optimizing appliance usage and reducing unnecessary energy consumption, contributing to a smaller carbon footprint.

The app also supports mental health by simplifying the cooking process, reducing the stress of coordinating multiple dishes, and breaking tasks into manageable steps. Users can focus on enjoying their gatherings without feeling overwhelmed.

Our AI is designed ethically, prioritizing user privacy and data security while ensuring fairness and inclusivity. The app is accessible to users of all skill levels and accommodates diverse cultural cooking practices and dietary needs. Customizable options allow users to align their cooking with personal values, such as prioritizing sustainable or low-energy cooking methods.

This project is not only ideal for individuals cooking for their family and friends but also has the potential to scale for larger applications. It can be a valuable tool for catering companies to streamline planning, restaurant chef teams to coordinate efforts seamlessly, and cookbook creators to organize and optimize recipes for diverse audiences. By accommodating both individual and professional users, the project stands to make a significant impact across various segments of the culinary industry.

Literally Cooking isn’t just a cooking assistant—it’s a thoughtful, sustainable, and health-conscious tool designed to make hosting easier and more enjoyable for everyone.

How I built it

We used Figma to design Literally Cooking and Flutter for our frontend. For our backend, we used Spoonacular API to gather recipe data, python and boto3 to call Bedrock/Converse, and trained a Bedrock model to understand how to create a mega recipe. For the Bedrock models, we used Claude 3 Sonnet for the prompt engineering a firebase to build up the knowledge base. We created an algorithm to have users score AI-outputted "mega-recipes" so that the AI could get better at returning desired outputs. We used Flask and Firebase to connect the frontend and backend to allow the model to reference a knowledge base based on data we had from our api.

Challenges I ran into

Training the model took a lot of trial and error to get it to have the correct output. When training our Bedrock model for the knowledge base, we realized that there was a lot of frontend and backend communication that we had to parse and send in different ways, and that training an AI model is not as simple as just telling it what to do.

Accomplishments that I'm proud of

We are proud of developing an accurate model that generates a "mega recipe" after multiple iterations of trial and error, and also of the process it took to train the model and have it learn from user input. We were not familiar with AWS AI technologies before the hackathon and were proud to have learned how to apply the knowledge we learned to our application. The set up of our frontend went smoothly. We are also proud of the fact that we put together a lot of different components, some of them which members of our team weren't very familiar with.

What I learned

We learned how to teach an AI model and have it reference a knowledge base in order to help when the output of the AI isn't good enough. We learned that when giving prompts to AI to generate a specific output, we have to be very detailed and consider almost every possible edge case that could occur, which pushed us to think about our use cases more deeply. We learned how to connect a knowledge base in the backend to a flutter frontend using flask and firebase.

What's next for Literally Cooking

Adjusting model settings further for even more accuracy of results. Adding computer vision to let people upload photos of food which the AI would then generate a recipe for. Adding a timer integrated in the app to remind you when items are done cooking. Giving users the freedom to add recipes to our database to increase customization and diversity. Combining more than two recipes at a time.

Built With

Share this project:

Updates