Fridgit

Inspiration

 30-40% of food in the United States gets wasted. That's ** ~$1500** per family per year, straight to the trash. We kept finding expired ingredients in the back of our fridges. For instance, milk that expired 2 weeks ago or veggies that became rotten because we forgot that our parents had bought for us to consume and enjoy. If AI can identify a particular face in a large crowd, then why can't we utilize it to track yogurt expiration dates? So we built Fridgit. A revolutionary application that uses AI to track expiration dates on items in an individual's refrigerator and provide them with recipes that they could try using ingredients that are near-expiration.

What it does

Fridgit uses AI to help families waste less food, save money, and shop smarter. Simply, point your camera at fridge items, and our application will identify products and their expiration dates from labels. These items are automatically logged into our virtual fridge with freshness timers. Push notifications such as "Your spinach expires in 2 days" or "Your Dosa Batter expires in 7 days" are instantly sent out. Additionally, Fridgit recommends recipes based on what's expiring soonest so a person can use the food items before it goes bad.

How we built it

We built Fridgit entirely on Replit (free version) for full-stack development and real-time collaboration during the hackathon. An API key from OpenAI was also used to implement an AI bot inside of our app, so that it could scan and provide recipes. Fridgit is currently a prototype, but it proves that our core idea works.

Challenges we ran into

The biggest technical hurdle was getting the AI bot to read messy expiration dates. Labels are small, blurry, or written in weird formats, so we had to train our model on fifty real fridge photos. We collected fridge photos from families with different cultural backgrounds to ensure that the AI worked for more than just one "standard" type of fridge. We also realized that people won't scan every single item, or that the app may not be able to read every item, so we added a feature that allows for one to quickly add that item or items.

Accomplishments that we're proud of

We tried to solve notification fatigue by reducing the amount of alerts to "ONE" smart ping per day, unless the user chooses to opt into "Expiry Alerts," which in this situation, they will receive the necessary amount of alerts. Additionally, we turned a "universal" problem into a one-tap solution. No one wants another "chore" app, so we made Fridgit feel automatic and accessible.

As a bonus, we had our parents test out the app and they didn't delete it, which felt like the real test of product market fit!

What we learned

Real fridges are NOTHING, and we repeat, NOTHING like stock photos. We learned we had to train on messy, real-world photos from diverse kitchens to make the AI-bot actually useful, so it could generate recipes using ingredients/items that are close to expiring.

What's next for Fridgit

People, oftentimes, take fruits and vegetables out of their original plastic packaging, so it would be highly beneficial to provide a "Smart Produce" tracking feature. Users would be able to take a photo of loose items using a second in-app camera scanner. The app will then use computer vision and AI image analysis to identify the produce type and estimate its ripeness based on visual cues. Additionally, we plan to build grocery integration to auto-add items from Instacart or Walmart receipts so scanning isn't always needed.

In the long term, we would like to partner with retailers to alert everyday people about discounts on items that will expire soon, which will save them money and keep food out of landfills. Eventually, Fridgit becomes a kitchen OS that helps plan meals and tracks nutrition. If we are able to do so, we would like to save at least 1.2 million pounds of food in year ONE, and change how a generation thinks and goes about waste.