What-is-for-dinner?

• 

  This is how my application Looks

Inspiration

This application is an ode to my bachelor days when i was working in Pune, India. I love to cook and eat tasty recipes but due to staying away from home, i was not able to get good home cooked food. Even outside food did not have that taste. So I would always call home (Lucknow, India) to my parents and get the recipes then shop for ingredients and then finally cook. Sometimes it was an error in my cooking and sometimes it was a miss in the recipe itself and sometimes I could not get the necessary ingredients. So those days were always an experiment with unknown results. At that time, not many people had food blogs or online cooking websites where the ingredient lists were easily available. Now this was at the back of my mind that if there was an application that could reverse-engineer the recipes from the ingredients that would be amazing.

What it does

The application asks for Raw ingredient list available in the kitchen. Then it generates two things. First are some of the recipes which can be made with these ingredients along with some accompaniments. For health conscious people, i have also included the calorie count per serving. Next it generates an image as to how the food should look. As a final response, i have integrated a chatbot at the end for people to ask clarifying questions.

How we built it

I created it from Scratch and used to think what it should look like. First i understood that for it to be an application, it should be fast and not have too many prompts. So there is just one prompt asking for user input for raw ingredients and the other two widgets generate Recipes and one more widget shows how it will look.

Challenges we ran into

I was not sure about how to use the widgets and the advanced filters used in different picture generation LLM models. I must have toggled and saved the filters at least 20times to get the exact values. Secondly i had set the values too high for the image generation so it was not coming correctly and in different colors and tones. Unable to generate more than one image.

Accomplishments that we're proud of

I was able to ask the right questions essentially use correct prompt engineering words to generate the correct outputs. It took me a while but first i was just getting one recipe. Then i toggled the prompt so it would generate all the recipes.

What we learned

I was able to generate all the recipes but i found that it is NOT possible as of now to generate more than ONE image for one prompt.

What's next for What-is-for-dinner?

I have used Amazon Q and i want to integrate the LLM with the traditional Awadhi recipes and family recipes so the data output which i am getting is authentic home-cooked or traditional regional recipes. I will definitely try to integrate that with my current application.

Built With

• bedrock