Foodify

Landing Page
Starting recipe finding
Recipe found!

Inspiration

As college students who don't have much time to mealplan multiple times a day, Foodify was started with a simple goal: simplify meal prep and provide recipes for nutritious, easy-to-make dishes. If you're indecisive, don't worry! Foodify asks the user questions to discern exactly what cuisine they're craving. You can even provide Foodify with the ingredients you have on hand, and it will effortlessly handcraft a recipe just for you!

What it does

Foodify simplifies your food search! This Google Gemini-powered chatbot streamlines recipe searches, offering personalized suggestions and tips. Easily navigate back and forth to explore diverse options, tailored to your tastes and dietary needs. Say hello to effortless culinary inspiration!

How we built it

First started with a boilerplate tutorial hosted on Google Colab, we were able to experiment with Google's Gemini AI with their vision and 1.5 Pro response models. This allowed us to fine-tune our model while progress was being made on setting up the base that would hold the model and connect it to our chatbot flow. The large amount of tinkering allowed us to hone in on an effective chatbot persona that would be able to consistently provide results that we wanted. Alongside this, we were able to build a pure python fullstack webapp through the use of the Reflex framework. By walking through our thought process and looking at issues alongside the Reflex crew, we were able to help Reflex identify potential issues that may arise in future production models and slowly fix bugs in our own code at the same time. Once both sides were tuned and completed, the integration of these parts required a few tweaks, but through trial-and-error we were able to connect the model with the chat-flow.

Challenges we ran into

Lack of API support for Amazon/Instacart cart creation led to an unsuccessful rabbithole. While it was possible to potentially create our own API for this task, writing all of the get- and post- endpoints would have taken too much time and diminished the quality of the rest of the project.

Accomplishments that we're proud of

Krish: working through multiple Github issues and coordinating with the team to make a product we can be proud of! Also spamming the Gemini AI with "interesting" personas to see funny responses whenever we got bored :)

Jason: being able to train and manipulate our AI persona to accurately and concisely provide consistent output to the user was an arduous process that required many iterations; reaching a stable scenario was extremely rewarding. This project was a huge step up from my first hackathon, which consisted of a simple python script.

Krishna: Creating a web app using Reflex and integrating Gemini API's LLM training is both fulfilling and impactful. By giving our bot a persona and specific goals, we enhance user attention and provide tailored food choices."

Harsh: Exploring the concept of Foodify, an innovative platform aimed at revolutionizing the culinary experience, while delving deeper into the capabilities of Google's Gemini API for training. This endeavor seeks to refine and enhance the AI chatbot, ensuring its accuracy and effectiveness in delivering personalized interactions and valuable assistance to users.

What we learned

We got to learn the intricacies of GitHub on a much deeper level, through struggling with merge conflicts as we first set up our project in figuring out how useful branches could become. GitHub allowed us to efficiently collaborate with teammates, without interfering with each others' work. Through this, we learned how to work even better as a team.

What's next for Foodify

Integrate an automated recipe-to-online food cart order. This could either be hosted on Amazon, Instacart, or other similar companies or brands. We would have to find a third-party API that allows us to connect with these grocery services, as we currently do not have an easily accessible way to connect our chat recipe finder to sources like these.