Authors

Inspiration

Jillian and Renata worked in restaurants for years and have personally thrown away buckets and buckets of food. Seeing the effect of COVID-19 on the service industry, inspired us to adapt our solution to today's circumstances. We came to the hackathon to push ourselves to contribute as much as we can during this time

What it does

We suggest how much food restaurants should be ordering. To do this we break down restaurant’s menu and use machine learning to predict future sales.

How we built it

The two methods tested for this were selected because they are methods that are often used for weather predictions. These are the Recurrent Neural Network (RNN) and Auto Regressive Integrated Moving Average (ARIMA) Model. Each of these models allow for previous outputs to be used as inputs. The RNN was built using inputs of the total sales per-day where the time of the year was indicated as well as local weather data to see how that affects sales.  ARIMA simply uses the past sales of a product to predict it’s future sales. The drawbacks with the RNN is that it is computationally slow, so we did not have time in this Hackathon to get accurate results. This neural network is also better at predicting when data is sequential which is not necessarily the case in restaurant data. The ARIMA models predict with a 86% accuracy in our tests with Icelandic restaurant data before the COVID-19 crisis.

In the past 48 hours we used Figma to make adjustments to our UX/UI design, LinkedIn to contact 20 restaurant owners in Denmark (we got positive feedback from the cowoner of OLIOLI and our Linked in requests were accepted by 11 co-owners or restaurant CEOs), and WebFlow to build a live website.

Challenges we ran into

We have so many ideas we want to put into our submission, we brainstormed about including some too good to go feature in the case our algorithm still has some food waste (less than before of course). We spoke to mentors about how in the future we can scale our solution and have some sort of aggregate  location based forecast if we could amass large amounts of data. Acquiring data is also difficult, we wanted to try and test our algorithms with Danish data, but we were not able to get a data set from a Danish restaurant.

Accomplishments that we're proud of

We are proud to have put together tangible deliverables and discuss our ideas without having met in person. It's really cool that technology today allows us to adapt to physical distancing requirements.

What we learned

How to go from Figma to a live website, how to manage a team remotely, how to adapt go to market strategies to strange circumstances and that market research takes a lot of time, effort and shamelessly reaching our to people.

What's next for GreenBytes

We really believe we have a solution that can help restaurants survive COVID-19 and thrive after it has passed. In order to implement our solution on a large scale we nee to continue refining our algorithms, set up a cloud based database (maybe in AWS) to take in larger amount of data, continue developing our website, and continue validating our product-market fit with restaurant owners.

Try It out

Hackathons

Technologies

figma, html, python, webflow

Devpost Software Identifier

257900