GPTasty Meals

Students have the time to cook. What they don't have, is the time to think about WHAT they should cook. That's where GPTasty Meals comes into play. List your pantry items, and get a custom recipe!

Comment

Inspiration

I use ChatGPT only for figuring out what I should be cooking for lunch/dinner. Which is ironic, considering I'm a student. A lot of the recipes I would find online would contain uncommon ingredients like Worcestershire sauce, or Water Chestnuts. As soon as I'd see this, I'd run to a different recipe. So I decided to conduct a market survey and figure out whether other students also face a similar problem.

What it does

It uses RAG to enhance the results that are generated by the LLM. It takes in an input of ingredients in your pantry, and uses that to generate a recipe for you, making sure there's no other ingredients that you don't have at hand.

How I built it

React frontend, express backend, and a MongoDB data store. I also used LlamaIndex for managing the RAG knowledge base, which integrates into an OpenAI LLM API call.

Challenges I ran into

The lack of a team. As soon as I realized that I may have to go solo, I needed to prioritize building of the app. I had to cut off some features, and make sure that I have an MVP that's up and running. Integrating the RAG model was also super complex, since I didn't understand anything.

Accomplishments that I'm proud of

  • I learned an entirely new concept - Retrieval Augmented Generation (RAG) in under 24 hours.
  • The entire core backend of the project works.
  • I got stuck 100 times. Got unblocked 99 times.

What I learned

  • This is definitely a team effort.
  • I wanted to give up. But there's something inside me that doesn't let me do that.
  • What makes Stanford Stanford. Just listening to other students talk about their ideas, it felt amazing. I felt extremely out of place, but it was magical.

What's next for GPTasty Meals

  • A much more refined RAG infusion, complete with a vector store.
  • The ability to adjust portion sizes based on the person's fitness activity levels, via a health/smartwatch API
  • Users can rate the recipes they made. The app will generate a flavor profile, and adjust future recipes according to that. This might employ the use of AI Agents.

Built With

Share this project:

Updates