Image to Recipe: Reducing Food Waste, Cost, Carbon Footprint

Prompting
,
1

hey Guys,

I am absolute noob to the world of AI, except for using Chat GPT’s frontend service to learn and do amazing things. Thanks to its capabilities I am now a bit better at assembling software, than I was before.

I am a food tech guy - and love to explore various aspects of both. One of the area that interests me is Environment and I feel it needs immediate attention. I did build a simple website for it, however have not come too far > https://eatlo.org/

Currently I am exploring possibilities of making an app, which would try to read the ingredients from the image and then generate a recipe or take some other suitable action based on that. The idea is users can put everything they have which is nearing expiration date, or they are not sure what to do - put everything in a picture, send it to app - and app will help you use them in best possible way as per your needs and requirements - hopefully reducing food waste, cost, carbon footprint etc in the process.

On the first go, it does sound simple - like just uploading images to chat GPT and asking it do do things, but apparently designing a proper flow for the above ‘prompt’ is essential to how the app behaves.

  1. For starters, I am wondering the 4o model - is it ok to send an image to it, ask it to identify ingredients and and then write a recipe accordingly?

  2. Should this be handled in multiple prompts - like one for getting the ingredients, parsing them then confirming them with user - probably then making another request for the recipe?

  3. Or this can be done like in a conversational mode - where user starts a conversation, sends a picture, and then it works like how chatgpt frontend does - somewhere giving the option parse and save recipe once user is satisfied.

I know it all sounds vague, but being newbie to all the technical verbiage of the ‘backend’ of AI, hopefully I would be able to clarify my use case enough to get some valuable advice on the same.

2

Hi!
I suggest to go ahead and just try things. Most of your questions can be answered for your specific case by trial and error, before optimizing.
As a general rule of thumb: the less the model has to do per each step the more control you have over the output.

Lastly, and I am sure you have already done this, it’s time to start working with real humans instead of just the AI to figure out what’s really needed and how your users get value from this concept.

1 Like