Inspiration

The idea for Food Flow came from a personal experience that changed my perspective on food waste. One evening, I saw a bakery near my home throw away bags of fresh bread while, just across the street, a family was searching for food in a donation box. Witnessing this disconnect made me realize how much good food goes to waste while people go hungry. That moment inspired me to create a solution that could bridge this gap in any community.

What it does

FoodFlow serves as a flexible system where food donors and recipients are brought together through a self-running, user-friendly, and automated algorithm—let's say, an app that connects customers in need of a meal with restaurants that have food that will go to waste. The users go to our marketplace and the donors will post food pictures and quantity and state their preferred time and place for pickup, while the recipients will browse the items that match their needs in the marketplace. Our geolocation technology ensures finding the right matches between donors and recipients and the use of AI to check food quality as well as making the best matching recommendations, and lastly, the inclusion of weather data for the securing of reliable deliveries. Green posts are identified for urgent attention in which a tight-knit group of people are managed in such a way by the system that they instantly respond to the needs of the community while also reducing differences in food access.

Accomplishments that we're proud of

We learned that technology can bring people together to solve real problems, but listening to the community and adapting to their needs is key. Every shared meal is a reminder of the impact small actions can have. The creation of FoodFlow helped us realize that the fundamental basis of any technical solution should be equity. We got to know how AI could be useful in solving intricate logistics issues, as well as, how AI could involve some constraints when the training data is not diverse and does not reflect the real-world complexities. It was realization for us when we brought into practice the biweekly development sprints gathering feedback directly from the communities we serve ensuring the technology remained undeniably current and inclusive. But the most significant thing that we came to acknowledge was that the existence of hunger did not mean the supply was a problem only—it was the issue of justice that, through creation of poverty and health disparities, had a more significant impact on the communities affected, who have also been marginalized.

What's next for Food-flow

We plan to expand Food Flow to more neighborhoods and partner with larger food networks. Our goal is to make food insecurity a thing of the past, one community at a time.

How we build it

The React Native frontend with Expo Router and Tailwind CSS for creating a clean interface was our way of implementing the frontend. Python and its AI-based image classification and weather response pipelines have become the powerful tools for the backend. To make the right pairings of donor and recipient for future blood transfusions, we employ KNN as a recommendation system that is a classic of its kind. The spoilage detection model, based on the MobileNetV2 architecture and trained on a set of 3,000+ images from 8 food categories and 6 spoilage levels, was developed by us to be used for the quality assessment of the food products. To cover data augmentation and hyperparameter tuning, the accuracy of 93.7% was achieved in the testing process through the weather service's API. When the weather situation changes, the weather forecasting pipeline gets data from the OpenWeatherMap and the NOAA for scoring the points, and doing it faster, the method of DBSCAN geospatial clustering with Dijkstra's shortest path works well. On the other hand, we have been working for the user and transaction data storage in Firestore, the AI model evaluations in Cloud Functions, and structured relational data in Supabase. After that, the image is sent to Cloudinary where a FastAPI backend is responsible for predictions. Apart from the infrastructure of Google Cloud, the deployment of the system also involves features like scalability, fast response times, and even data security.

Built With

• aiml
• apis
• cloud
• css
• data
• dbscan
• dijkstra
• github
• google
• gps
• knn
• native
• noa
• openwhetermap
• python
• react
• react-native
• storage
• typescript