Pedible

Inspiration

Got a curious critter at home? Is your furry pal a foodie, tasting everything on the ground, and you're left scratching your head about what it's called or if it is safe? Or maybe you're eyeing that last bite of your snack, wondering if you can share it with your four-legged friend. Fear not! Pedible is here to save you from all your pet-related predicaments

Aren't we all worried when our pets sniff out mysterious flowers or plants on our walk? Or the irresistible puppy eyes they give us when we are munching on some fruit? But who wants to trawl through endless online paragraphs to find a nugget of info? That’s where Pedible comes in, saving the day and our sanity!

What it does

Pedible is a full-stack web application that allows users to take quick snapshots of an object and identifies whether the object is potentially harmful to your pets. It uses Computer Vision and Google Gemini to identify objects and infer from a single-shot image.

How we built it

We embark on the journey of experiments! For this project, we are using brand-new tech stacks and frameworks that none of us had previous experience with. We used reflex as our main tech stack for both frontend and backend. For the CV pipeline, we used Microsoft's BEiT model with weights pre-trained on ImageNet22K and the last classification layer set to first identity the object, then we passed the prediction to the amazing Google Gemini API which acts as both an ensemble method for object classification and as an inference, we have also parsed the Google Gemini API's responses into JSON and return the instance to frontend

Challenges we ran into

Our team ran into some challenges when it came to the initially choosing the techonolgy for our project. We wanted to work with Swift at first and deploy an iOS app, but only one of us had a Macbook, essentially making it almost impossible to work with it. It took some time, but we settled on the bold route and decided to use Reflex, one of the sponsored companies here at LAHacks, and their fully Python framework. Getting used to the syntax and the functions took awhile to udnerstand, but the front-end did a great job piecing together a prototype.

Accomplishments that we're proud of

Teamwork and communications. It started as two teammates going into it not knowing anyone else, but we found our third and we learned a lot from each other. We are also very proud of being able to tackle new challenges together, such as the technology we used, and it resulted in a nicely put-together WORKING Computer Vision/AI prototype. Being able to go out of our comfort zones made us feel like better programmers.