mealize

Inspiration

Meal planning is hard. Saving money? harder. Coming up with creative recipes? The hardest. Inspired by watching both the struggle of college students to both find cheap and easy meals to create out of limited amounts of options as well as one team member watching a roommate browse through coupon sites like Reebee and Flipp manually to find coupons and then go grocery shopping, we felt there needed to be a way that people could both save money, waste less food, and have more creative meals all in one smooth workflow.

What it does

The app flows like the following:

• detects nearby grocery stores
• scrapes Flipp for the items that are on sale at said grocery store
• asks for dietary restrictions
• uses a live camera stream which recognizes a picture you take (of a food item currently in your fridge) and deploys a machine learning model to classify it
• generates recipes from existing and items on sale!

How we built it

This app was built with Firebase ML Kit, Android Studio and Google Cloud Platform Vision Edge API. As well, we used HTTP requests to do tons of search querying for various filters. The UI/UX of the app was originally designed on Figma.

Challenges we ran into

A lot of the backend of this application involves parsing strings, cutting strings, matching strings, concatenating strings - etc! One pretty specific but pretty significant challenge we ran into was properly matching on sale items with recipes from a database. For instance, grocery stores such as Walmart would also have shampoo on sale, but shampoo can't be used as a food ingredient! We need to know to ignore it. Further more, strawberries might be on sale but the recipe had required berry jam...would this still be a match? These issues were solved by parsing for popularity, checking against multiple references, and reading up on tons of documentation about how people have solved similar issues.

Accomplishments that we're proud of

• An original project. It was truly a project inspired by a problem we faced daily and so we developed an application that attempted to solve it.
• Beautiful and seamless UI
• Working prototype
• Completely implemented backend of Google Cloud machine learning model in Android Studio
• Able to generate around 25 unique recipes for a given instance of a user walking through our app (this was tested by setting different geo-coordinate locations all over Toronto).

What we learned

Java can become very, very object oriented.

What's next for mealize

• Train the machine learning model on GCP with a greater variety of images.
• Implement even more inclusive tags for different dietary restrictions
• Be an app that can be shared with our friends to ACTUALLY use on a daily basis (because it is truly a pain point all of us share)
• (Very far future!) Potentially utilize reinforcement learning to eventually learn of a user's preferences in recipes?

Built With

• android-studio
• cloud-automl
• figma
• firebase-ml-kit
• google-cloud
• java