What it does
We created an administrative tool to manage the current stock of the products and anticipate the available resources into the near future, warning the user before getting out of stock based on provisions that were made thanks to the available data.
It also allows us to see the evolution of the sales from a specific model.
How we built it
In the beginning we used powerful tools like BigQuery, which uses Spanner in the background, to generate and manipulate the data fastly, creating more than 19 million rows of different data, like sales or stock.
We connected the data to a FastAPI backend, which connects to our front-end developed with Vue.js.
The backend handles the data previously analyzed on BigQuery to provide the important information to the customer, like a provision of the next models that will arrive.
We also started to train a Long Short-Term Memory neural network, but due the random data, as well as some technical issues, for the moment our backend is returning a provision based on the average of previous days.
Challenges we ran into
Due to security issues, one of our Google Cloud projects was hacked during the hackathon and someone wasted like $1500 USD on computing.
Also the network issue affected us serverously, since the connection between client and server was restricted during some hours.
Accomplishments that we're proud of
We learned a lot of different technologies during our project. We also are proud about our improvement, since for two of us this is our first hackathon.
What we learned
We learned theory and how to create an AI using PyTorch. Also we learned how to work with git.
We also learnt some basics about Data Analysis and how to work with cloud computing tools or how to configure the firewall to deploy our project outside localhost.
Last but not the least, we also started to learn how to use VueJs and other frontend stuff.
What's next for MotomAI
We would like to add more information on our frontend, already available on the Backend.
Additionally, we would like to replace the current estimation based on the average for a LSTM model and polish some performance issues.