The team has worked in restaurants for years and noticed day after day a large amount of food ended up in the garbage daily. The more we learned about food waste and its negative effects on the planet, water, land usage, and biodiversity we knew something had to be done.

What it does

We offer machine-learning-based food ordering suggestions for restaurants on how much food to order to reduce food waste. We have developed a progressive web application that breaks down menus, tracks inventory, and predicts future food consumption. More specifically we use a recurrent neural network that inputs past sales, weather data like temperature, wind speed, snow, rain, and sunlight hours, and COVID-19 statistics like the number of cases per day and rate of infection to understand the relationship between these factors and future food consumption to predict future sales.

Once we predict future sales, we suggest how much of each raw ingredient the restaurant should order in the upcoming days. If the restaurant agrees with our suggestion, they can approve the order and we will automatically send out the order to all of their distributors.

How I built it

We input 9 months of sales data, from a local juice bar in Iceland called Lemon. We also input mean temperature, max wind speed, day of the week, snow depth, precipitation, sunlight hours, day of the week, and the number of COVID-19 cases in Iceland. The weather data was retrieved from and the data about COVID-19 in Iceland was retrieved from A recurrent neural network was trained using Tensor flow for Python on all but three days of the given data, the last three days were used as validation to evaluate the accuracy of our models.

Challenges I ran into

We had a lot of discussion about how to account for COVID-19 and the effect it has on restaurants. Worldwide we know restaurants have been going out of business due to the pandemic. We know for our solution to be helpful we have to reduce operational costs while accounting for how the pandemic affects restaurant sales. We noticed that on the Icelandic website the COVID-19 statistics could be downloaded as CSV files which is a compatible format for our models. Another challenge we ran into was processing so much data. Just cleaning and pre-processing the data before the algorithm was trained took a very long time, and at one point, we had to stop running the program because the computer was so loud!

Accomplishments that I'm proud of

We asked the restaurant to provide us with data about how much food they actually purchased. They told us how much food they ordered for a week in January, we compared how much food we would have suggested to what they ordered, and actual sales. We found we could have reduced around 2.000 kilos of food waste in one month. Ultimately we decided to run the model we submitted from February because that is when COVID-19 began to affect Iceland. As we are in the midst of another wave of the pandemic, it is evident we need to adapt our solution to work even while the influence of the Corona virus fluctuates and persists.

What I learned

To be flexible and ask for help. During this weekend some things didn't go to plan and we would have to try something again and make some tweaks. We were also connected with an amazing network that took the time to answer technical questions and read over iterations of our pitch.

What's next for GreenBytes

Our next steps are to further analyze our algorithms, conduct LCA analyses, and run full-service tests in restaurants. We would like to involve University students so we can produce better results and encourage people with strong technical backgrounds to pursue green tech and implement solutions for the planet.

Share this project: